Speechmatics growth playbook

Render the cookie banner as light HTML before Cookiebot loads, load marketing pixels only after consent and after the page is interactive, remove the tags no one reports on, split the homepage JavaScript so the live transcription demo loads on click, and set a performance budget checked on every Vercel deploy.

Developers judge an API company by its site, and the self-serve funnel starts on this page; a faster page lifts sign-ups and stops the site from contradicting the latency pitch.

[ ] Owner named for each third-party service
[ ] Unused tags removed
[ ] Cookie banner rendered as static HTML
[ ] Ad and social pixels after consent and interaction
[ ] Live demo JavaScript loaded on click
[ ] JavaScript budget enforced on deploy
[ ] Field INP under 200 ms

• Losing attribution data

  Keep the tags that feed reports, loaded after consent.

• Consent compliance

  Have legal review the banner change.

• Google PageSpeedHomepage, tested October 22, 2026: mobile performance 45/100 (desktop 57); in the phone lab test the largest element takes 11.8 seconds to appear and it is the cookie-consent text; total blocking time 692 ms. Real Chrome visitors fail Core Web Vitals on responsiveness (INP 375 ms; the "good" limit is 200 ms), while their load time passes (LCP 2.15 s). The page makes 163 requests with 1.76 MB of JavaScript, about 637 KB of it unused, and loads 16 third-party services (about 1 MB), including Google Tag Manager (545 KB), Cookiebot (204 KB), X/Twitter (132 KB), Clarity, Reddit, Bing Ads and a Spotify pixel.Source
• HiringGrowth Engineer (Belgrade, posted October 1, 2026): owns "the backend development and ongoing health of the Speechmatics website and self-serve portal", Vercel deployments, Python queries on the self-serve portal, "opportunities to improve conversion across the self-serve funnel", and "the ongoing improvement of site performance, loading speed, and technical reliability".Source
• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source

Generate the sitemap from Contentful for every published case study and use case, exclude preview, test and thank-you pages (and noindex them), add Article and Organization schema to case studies, and resubmit in Search Console.

Case studies are the pages that win buyer searches such as "speech to text for captioning" and are what AI assistants quote; they should be the first pages search engines find.

• Other content types missing

  Compare every Contentful type with the sitemap.

• Your websiteSitemap (479 URLs, read October 22, 2026): the case-studies hub links 11 case studies, but the sitemap lists only 4 case-study pages; 10 of the hub's 11 are missing, including NCI, AI-Media, Vail Systems/Prosodica, Media Track, Edvak EHR, Callers, Limecraft, ENCO, Liquid Voice and Tools On Air, although they return 200 with self-canonicals. Use-case pages for EdTech and media monitoring, linked from the homepage, are missing too. The sitemap does list a /component-preview page and /component-preview/accordion (which returns 404) and /thank-you. The site also publishes /llms.txt and /llms-full.txt.Source
• HiringWebsite Coordinator (Belgrade, posted October 20, 2026): to own the website, "how it looks, what it says, and how easily people (and AI search tool and agents) find it"; builds pages in Contentful, monitors Google Search Console and optimises "metadata, headings, internal linking, content structure, and schema to boost visibility in both traditional and AI-driven search".Source
• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source

Detect the concurrency limit in the portal and API response, return a clear message with a one-click upgrade link, email the account owner the same day, and show in the portal how close each project is to the limit.

The limit marks a developer who has moved from testing to real traffic: the best moment to convert.

• Feels pushy

  Offer a short trial of higher concurrency instead of a hard sell.

• Your websitePricing page: title "Speech to Text from $0.121/hr", heading "from $0.12/hr"; "Start free • $100 in credits No card required"; models Melia 1 ($0.12/hr, 56 languages, switches language mid-sentence), Enhanced ($0.38/hr), Oak 1 Medical ($0.15/hr) and Linden 1 agent STT (real-time only, built-in turn detection); real-time concurrency "50 (2 on Free)"; SSO a "Paid add-on, $280/mo"; on-prem containers, virtual appliance and on-device for Enterprise only; a pricing calculator.Source
• HiringGrowth Engineer (Belgrade, posted October 1, 2026): owns "the backend development and ongoing health of the Speechmatics website and self-serve portal", Vercel deployments, Python queries on the self-serve portal, "opportunities to improve conversion across the self-serve funnel", and "the ongoing improvement of site performance, loading speed, and technical reliability".Source
• HiringDeveloper Relations Advocate (San Francisco, posted October 22, 2026, $100,000-$140,000): "Be the face of Speechmatics in San Francisco ... bringing back the market insight that shapes our US growth"; "Get developers from first API call to working product"; webinars, hackathons, meet-ups, tutorials and code samples; "Use community and content data to show what's working".Source

Add three questions (live or recorded, one language or mixed, medical or not) that recommend a model, show its price and link to a code sample. Make the title and heading show the same price.

A clear starting choice reduces drop-off between pricing and the first API call.

• Wrong recommendation

  Let users switch models freely with credits.

• Your websitePricing page: title "Speech to Text from $0.121/hr", heading "from $0.12/hr"; "Start free • $100 in credits No card required"; models Melia 1 ($0.12/hr, 56 languages, switches language mid-sentence), Enhanced ($0.38/hr), Oak 1 Medical ($0.15/hr) and Linden 1 agent STT (real-time only, built-in turn detection); real-time concurrency "50 (2 on Free)"; SSO a "Paid add-on, $280/mo"; on-prem containers, virtual appliance and on-device for Enterprise only; a pricing calculator.Source
• HiringSenior Software Engineer - Front-End (London and Cambridge, posted September 25, 2026): "we're expanding our user platform offering"; designs and runs A/B tests, uses Mixpanel, and runs user interviews "to understand how people use the product and what drives adoption".Source

Add a case-study block to each use-case page using a Contentful reference field, and a use-case filter on the case-studies hub. Where a use case has no story (for example dental or veterinary), mark it for the next customer interview.

Proof next to the claim lifts conversion, and gaps show where to find the next story.

• Few stories for some uses

  Use short quotes until full stories exist.

• Your websiteSitemap (479 URLs, read October 22, 2026): the case-studies hub links 11 case studies, but the sitemap lists only 4 case-study pages; 10 of the hub's 11 are missing, including NCI, AI-Media, Vail Systems/Prosodica, Media Track, Edvak EHR, Callers, Limecraft, ENCO, Liquid Voice and Tools On Air, although they return 200 with self-canonicals. Use-case pages for EdTech and media monitoring, linked from the homepage, are missing too. The sitemap does list a /component-preview page and /component-preview/accordion (which returns 404) and /thank-you. The site also publishes /llms.txt and /llms-full.txt.Source
• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source
• Your websiteStartup Program: "Up to $50k in credits", "Technical guidance", "Monthly office hours with our engineers", with a first cohort shown. On-device has its own page with a qualification form. Use cases include AI voice agents, medical, dental, veterinary and legal transcription, contact centers, meeting platforms and captioning.Source

Enrich each new portal account by email domain, flag companies in the target segments or above a usage threshold, and send them to a BDR with their usage summary (model, language, real-time or batch). The BDR offers a Solutions Engineer session, not a sales call.

Engineers already testing the API are the warmest enterprise leads, and a technical offer fits how they buy.

• Developers dislike sales contact

  Lead with technical help and an easy opt-out.

• HiringGo-to-market hiring: Solutions Engineers in Belgrade (posted September 11, 2026) and Mexico City (posted September 29) to run discovery, demos and proofs of concept in "cloud or on-prem setups" and support conferences; Senior Business Development Representatives in London and Cambridge for "enterprise and mid-market accounts" with "technical buyers (e.g. CTOs)".Source
• Your websitePricing page: title "Speech to Text from $0.121/hr", heading "from $0.12/hr"; "Start free • $100 in credits No card required"; models Melia 1 ($0.12/hr, 56 languages, switches language mid-sentence), Enhanced ($0.38/hr), Oak 1 Medical ($0.15/hr) and Linden 1 agent STT (real-time only, built-in turn detection); real-time concurrency "50 (2 on Free)"; SSO a "Paid add-on, $280/mo"; on-prem containers, virtual appliance and on-device for Enterprise only; a pricing calculator.Source
• HiringGrowth Engineer (Belgrade, posted October 1, 2026): owns "the backend development and ongoing health of the Speechmatics website and self-serve portal", Vercel deployments, Python queries on the self-serve portal, "opportunities to improve conversion across the self-serve funnel", and "the ongoing improvement of site performance, loading speed, and technical reliability".Source

Run a monthly build night where developers ship a voice agent in two hours using Linden with LiveKit or Pipecat, with free credits, a prize and a demo slot. Publish each winning project as a sample and record short clips.

Hands-on events create first API calls on the spot and community samples that keep working after the event.

• Low turnout

  Co-host with an established meet-up.

• HiringDeveloper Relations Advocate (San Francisco, posted October 22, 2026, $100,000-$140,000): "Be the face of Speechmatics in San Francisco ... bringing back the market insight that shapes our US growth"; "Get developers from first API call to working product"; webinars, hackathons, meet-ups, tutorials and code samples; "Use community and content data to show what's working".Source
• Developer ecosystemDeveloper reach: 14 npm packages with about 47,700 weekly downloads (@speechmatics/real-time-client about 18,900); 55 public GitHub repos, the most-starred "speechmatics-academy" (164 stars); a Discord community; docs on Docusaurus; YouTube about 2,220 subscribers, 256 videos and 5.16 million views; LiveKit and Pipecat integrations, and Vapi and Pipecat among the logos on the site.Source
• Your websitePricing page: title "Speech to Text from $0.121/hr", heading "from $0.12/hr"; "Start free • $100 in credits No card required"; models Melia 1 ($0.12/hr, 56 languages, switches language mid-sentence), Enhanced ($0.38/hr), Oak 1 Medical ($0.15/hr) and Linden 1 agent STT (real-time only, built-in turn detection); real-time concurrency "50 (2 on Free)"; SSO a "Paid add-on, $280/mo"; on-prem containers, virtual appliance and on-device for Enterprise only; a pricing calculator.Source

Instrument sign-up, API key created, first request, first successful transcript and first real-time session in Mixpanel. Test a guided first run in the portal (upload a file or speak, see the code that did it) and trigger help emails when a step stalls.

Developers who reach a result in the first session convert and stay far more often.

• Over-instrumenting

  Track only the five steps first.

• HiringDeveloper Relations Advocate (San Francisco, posted October 22, 2026, $100,000-$140,000): "Be the face of Speechmatics in San Francisco ... bringing back the market insight that shapes our US growth"; "Get developers from first API call to working product"; webinars, hackathons, meet-ups, tutorials and code samples; "Use community and content data to show what's working".Source
• HiringSenior Software Engineer - Front-End (London and Cambridge, posted September 25, 2026): "we're expanding our user platform offering"; designs and runs A/B tests, uses Mixpanel, and runs user interviews "to understand how people use the product and what drives adoption".Source
• HiringGrowth Engineer (Belgrade, posted October 1, 2026): owns "the backend development and ongoing health of the Speechmatics website and self-serve portal", Vercel deployments, Python queries on the self-serve portal, "opportunities to improve conversion across the self-serve funnel", and "the ongoing improvement of site performance, loading speed, and technical reliability".Source
• Your websitePricing page: title "Speech to Text from $0.121/hr", heading "from $0.12/hr"; "Start free • $100 in credits No card required"; models Melia 1 ($0.12/hr, 56 languages, switches language mid-sentence), Enhanced ($0.38/hr), Oak 1 Medical ($0.15/hr) and Linden 1 agent STT (real-time only, built-in turn detection); real-time concurrency "50 (2 on Free)"; SSO a "Paid add-on, $280/mo"; on-prem containers, virtual appliance and on-device for Enterprise only; a pricing calculator.Source

Translate the homepage, pricing, contact-center and voice-agent pages and the Spanish and Portuguese language pages, with Spanish audio samples in the live demo and a Spanish-English code-switching sample. Route Latin American leads to the Mexico City engineer.

Local-language pages and demos in the buyer's own accent are rare among speech APIs and show the product's strength directly.

• Support in Spanish

  Start with sales and solutions engineering only.

• HiringGo-to-market hiring: Solutions Engineers in Belgrade (posted September 11, 2026) and Mexico City (posted September 29) to run discovery, demos and proofs of concept in "cloud or on-prem setups" and support conferences; Senior Business Development Representatives in London and Cambridge for "enterprise and mid-market accounts" with "technical buyers (e.g. CTOs)".Source
• Your websiteComparison pages: nine "alternative" pages (AssemblyAI, AWS, Deepgram, ElevenLabs Scribe, Gladia, Google, Microsoft Azure Speech, OpenAI, Soniox) plus a "How we compare" page. Language pages: about 60 /speech-to-text/<language> pages, including Spanish, Portuguese, Welsh, Irish, Basque and Swahili. The site itself is English only.Source
• Your websitePricing page: title "Speech to Text from $0.121/hr", heading "from $0.12/hr"; "Start free • $100 in credits No card required"; models Melia 1 ($0.12/hr, 56 languages, switches language mid-sentence), Enhanced ($0.38/hr), Oak 1 Medical ($0.15/hr) and Linden 1 agent STT (real-time only, built-in turn detection); real-time concurrency "50 (2 on Free)"; SSO a "Paid add-on, $280/mo"; on-prem containers, virtual appliance and on-device for Enterprise only; a pricing calculator.Source
• Competitor researchRivals' own homepages: Deepgram, "Enterprise Voice AI: STT, TTS & Agent APIs"; AssemblyAI, "AI models to transcribe and understand speech"; Gladia, "AI audio infrastructure for voice products"; Soniox, "realtime speech-to-text, text-to-speech, and translation in 60+ languages" with "seamless language switching".See the research Source

Publish one maintained template per framework (Linden STT plus turn detection, with optional TTS), deployable in one click, linked from docs, the voice agents page and each framework's own integration docs.

A template that works in five minutes wins the framework choice moment.

• Templates break with framework updates

  Run CI tests weekly.

• Developer ecosystemDeveloper reach: 14 npm packages with about 47,700 weekly downloads (@speechmatics/real-time-client about 18,900); 55 public GitHub repos, the most-starred "speechmatics-academy" (164 stars); a Discord community; docs on Docusaurus; YouTube about 2,220 subscribers, 256 videos and 5.16 million views; LiveKit and Pipecat integrations, and Vapi and Pipecat among the logos on the site.Source
• Your websitePricing page: title "Speech to Text from $0.121/hr", heading "from $0.12/hr"; "Start free • $100 in credits No card required"; models Melia 1 ($0.12/hr, 56 languages, switches language mid-sentence), Enhanced ($0.38/hr), Oak 1 Medical ($0.15/hr) and Linden 1 agent STT (real-time only, built-in turn detection); real-time concurrency "50 (2 on Free)"; SSO a "Paid add-on, $280/mo"; on-prem containers, virtual appliance and on-device for Enterprise only; a pricing calculator.Source
• HiringDeveloper Relations Advocate (San Francisco, posted October 22, 2026, $100,000-$140,000): "Be the face of Speechmatics in San Francisco ... bringing back the market insight that shapes our US growth"; "Get developers from first API call to working product"; webinars, hackathons, meet-ups, tutorials and code samples; "Use community and content data to show what's working".Source
• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source

Publish an Oak page with term-level accuracy results on a public clinical test set, a self-serve test where buyers upload their own de-identified audio, the HIPAA path, and the two medical stories. Give Solutions Engineers the same test for proofs of concept.

Letting clinical buyers measure accuracy on their own audio is more convincing than any claim.

• Uploading patient audio

  Require de-identified audio and state data handling clearly.

• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source
• News coverageHeadline on October 14, 2026: "Speechmatics wows US with new medical speech recognition model - and NVIDIA has a key role". Wikipedia and company data: founded 2006 in Cambridge, England; about 113 employees; CEO Katy Wigdahl.Source
• Your websitePricing page: title "Speech to Text from $0.121/hr", heading "from $0.12/hr"; "Start free • $100 in credits No card required"; models Melia 1 ($0.12/hr, 56 languages, switches language mid-sentence), Enhanced ($0.38/hr), Oak 1 Medical ($0.15/hr) and Linden 1 agent STT (real-time only, built-in turn detection); real-time concurrency "50 (2 on Free)"; SSO a "Paid add-on, $280/mo"; on-prem containers, virtual appliance and on-device for Enterprise only; a pricing calculator.Source
• Your websiteStartup Program: "Up to $50k in credits", "Technical guidance", "Monthly office hours with our engineers", with a first cohort shown. On-device has its own page with a qualification form. Use cases include AI voice agents, medical, dental, veterinary and legal transcription, contact centers, meeting platforms and captioning.Source

Build a kit: upload a sample of the customer's audio, run Speechmatics and up to two rivals the customer names, and produce a word-error report by speaker, accent and language, plus latency figures. Use it in every PoC and save the anonymised results.

A consistent, fast PoC shortens enterprise cycles and builds a library of real-world accuracy evidence.

• Rival terms of use

  Run rivals only with the customer's own accounts and consent.

• HiringGo-to-market hiring: Solutions Engineers in Belgrade (posted September 11, 2026) and Mexico City (posted September 29) to run discovery, demos and proofs of concept in "cloud or on-prem setups" and support conferences; Senior Business Development Representatives in London and Cambridge for "enterprise and mid-market accounts" with "technical buyers (e.g. CTOs)".Source
• Your websiteComparison pages: nine "alternative" pages (AssemblyAI, AWS, Deepgram, ElevenLabs Scribe, Gladia, Google, Microsoft Azure Speech, OpenAI, Soniox) plus a "How we compare" page. Language pages: about 60 /speech-to-text/<language> pages, including Spanish, Portuguese, Welsh, Irish, Basque and Swahili. The site itself is English only.Source
• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source

Publish a short profile of each cohort company, run a demo day with investors and press, give graduates a first-year discount, and track credits used and revenue after graduation.

Visible startup success attracts the next cohort and turns credits into paying customers.

• Credits used without conversion

  Check usage at mid-point and offer help.

• Your websiteStartup Program: "Up to $50k in credits", "Technical guidance", "Monthly office hours with our engineers", with a first cohort shown. On-device has its own page with a qualification form. Use cases include AI voice agents, medical, dental, veterinary and legal transcription, contact centers, meeting platforms and captioning.Source
• HiringDeveloper Relations Advocate (San Francisco, posted October 22, 2026, $100,000-$140,000): "Be the face of Speechmatics in San Francisco ... bringing back the market insight that shapes our US growth"; "Get developers from first API call to working product"; webinars, hackathons, meet-ups, tutorials and code samples; "Use community and content data to show what's working".Source
• Developer ecosystemDeveloper reach: 14 npm packages with about 47,700 weekly downloads (@speechmatics/real-time-client about 18,900); 55 public GitHub repos, the most-starred "speechmatics-academy" (164 stars); a Discord community; docs on Docusaurus; YouTube about 2,220 subscribers, 256 videos and 5.16 million views; LiveKit and Pipecat integrations, and Vapi and Pipecat among the logos on the site.Source

Pick 50 questions (by language and use case), check monthly which vendors assistants name and which pages they cite, and strengthen the language and use-case pages that lose: clear facts, accuracy figures, code samples and FAQ schema.

Being the named answer for a language is a lasting acquisition channel, and Speechmatics' language breadth is a real advantage.

• Answers vary

  Track trends over months.

• HiringWebsite Coordinator (Belgrade, posted October 20, 2026): to own the website, "how it looks, what it says, and how easily people (and AI search tool and agents) find it"; builds pages in Contentful, monitors Google Search Console and optimises "metadata, headings, internal linking, content structure, and schema to boost visibility in both traditional and AI-driven search".Source
• Your websiteComparison pages: nine "alternative" pages (AssemblyAI, AWS, Deepgram, ElevenLabs Scribe, Gladia, Google, Microsoft Azure Speech, OpenAI, Soniox) plus a "How we compare" page. Language pages: about 60 /speech-to-text/<language> pages, including Spanish, Portuguese, Welsh, Irish, Basque and Swahili. The site itself is English only.Source
• Your websiteSitemap (479 URLs, read October 22, 2026): the case-studies hub links 11 case studies, but the sitemap lists only 4 case-study pages; 10 of the hub's 11 are missing, including NCI, AI-Media, Vail Systems/Prosodica, Media Track, Edvak EHR, Callers, Limecraft, ENCO, Liquid Voice and Tools On Air, although they return 200 with self-canonicals. Use-case pages for EdTech and media monitoring, linked from the homepage, are missing too. The sitemap does list a /component-preview page and /component-preview/accordion (which returns 404) and /thank-you. The site also publishes /llms.txt and /llms-full.txt.Source

Publish an open-source harness on GitHub that runs any provider on public test sets or the developer's own audio, link it from each comparison page, and date every published result.

Inviting developers to check the numbers turns a claim into a test they run, and the harness spreads on GitHub.

• Rivals improve

  Refresh every quarter.

• Your websiteComparison pages: nine "alternative" pages (AssemblyAI, AWS, Deepgram, ElevenLabs Scribe, Gladia, Google, Microsoft Azure Speech, OpenAI, Soniox) plus a "How we compare" page. Language pages: about 60 /speech-to-text/<language> pages, including Spanish, Portuguese, Welsh, Irish, Basque and Swahili. The site itself is English only.Source
• Developer ecosystemDeveloper reach: 14 npm packages with about 47,700 weekly downloads (@speechmatics/real-time-client about 18,900); 55 public GitHub repos, the most-starred "speechmatics-academy" (164 stars); a Discord community; docs on Docusaurus; YouTube about 2,220 subscribers, 256 videos and 5.16 million views; LiveKit and Pipecat integrations, and Vapi and Pipecat among the logos on the site.Source
• Competitor researchRivals' own homepages: Deepgram, "Enterprise Voice AI: STT, TTS & Agent APIs"; AssemblyAI, "AI models to transcribe and understand speech"; Gladia, "AI audio infrastructure for voice products"; Soniox, "realtime speech-to-text, text-to-speech, and translation in 60+ languages" with "seamless language switching".See the research Source

Run all major speech APIs quarterly on public test sets by language and accent, publish the leaderboard with the method and harness, and let researchers submit test sets. Present results at developer conferences.

A trusted leaderboard becomes the reference developers and analysts cite, and it showcases where Speechmatics leads.

• Seen as biased

  Open method and outside reviewers.

• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source
• Your websiteComparison pages: nine "alternative" pages (AssemblyAI, AWS, Deepgram, ElevenLabs Scribe, Gladia, Google, Microsoft Azure Speech, OpenAI, Soniox) plus a "How we compare" page. Language pages: about 60 /speech-to-text/<language> pages, including Spanish, Portuguese, Welsh, Irish, Basque and Swahili. The site itself is English only.Source
• Competitor researchRivals' own homepages: Deepgram, "Enterprise Voice AI: STT, TTS & Agent APIs"; AssemblyAI, "AI models to transcribe and understand speech"; Gladia, "AI audio infrastructure for voice products"; Soniox, "realtime speech-to-text, text-to-speech, and translation in 60+ languages" with "seamless language switching".See the research Source
• Developer ecosystemDeveloper reach: 14 npm packages with about 47,700 weekly downloads (@speechmatics/real-time-client about 18,900); 55 public GitHub repos, the most-starred "speechmatics-academy" (164 stars); a Discord community; docs on Docusaurus; YouTube about 2,220 subscribers, 256 videos and 5.16 million views; LiveKit and Pipecat integrations, and Vapi and Pipecat among the logos on the site.Source

Offer a limited self-serve on-device SDK (one platform, a small set of languages, a usage-based licence) with a free trial, then hand larger deployments to sales. Track trial-to-licence conversion.

On-device is a clear difference from cloud-only rivals, and self-serve opens it to many smaller apps.

• Support load

  Limit to one platform first.

• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source
• Your websitePricing page: title "Speech to Text from $0.121/hr", heading "from $0.12/hr"; "Start free • $100 in credits No card required"; models Melia 1 ($0.12/hr, 56 languages, switches language mid-sentence), Enhanced ($0.38/hr), Oak 1 Medical ($0.15/hr) and Linden 1 agent STT (real-time only, built-in turn detection); real-time concurrency "50 (2 on Free)"; SSO a "Paid add-on, $280/mo"; on-prem containers, virtual appliance and on-device for Enterprise only; a pricing calculator.Source
• Your websiteStartup Program: "Up to $50k in credits", "Technical guidance", "Monthly office hours with our engineers", with a first cohort shown. On-device has its own page with a qualification form. Use cases include AI voice agents, medical, dental, veterinary and legal transcription, contact centers, meeting platforms and captioning.Source
• HiringSenior Software Engineer - Front-End (London and Cambridge, posted September 25, 2026): "we're expanding our user platform offering"; designs and runs A/B tests, uses Mixpanel, and runs user interviews "to understand how people use the product and what drives adoption".Source

Agree co-marketing with each platform: joint launch posts for new models (Linden, Oak), shared templates, a referral credit, and being listed as a recommended provider in their docs. Track sign-ups by partner.

One default setting in a popular framework brings more developers than any campaign.

• Partners stay neutral

  Win on templates and support, not exclusivity.

• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source
• Developer ecosystemDeveloper reach: 14 npm packages with about 47,700 weekly downloads (@speechmatics/real-time-client about 18,900); 55 public GitHub repos, the most-starred "speechmatics-academy" (164 stars); a Discord community; docs on Docusaurus; YouTube about 2,220 subscribers, 256 videos and 5.16 million views; LiveKit and Pipecat integrations, and Vapi and Pipecat among the logos on the site.Source
• Your websitePricing page: title "Speech to Text from $0.121/hr", heading "from $0.12/hr"; "Start free • $100 in credits No card required"; models Melia 1 ($0.12/hr, 56 languages, switches language mid-sentence), Enhanced ($0.38/hr), Oak 1 Medical ($0.15/hr) and Linden 1 agent STT (real-time only, built-in turn detection); real-time concurrency "50 (2 on Free)"; SSO a "Paid add-on, $280/mo"; on-prem containers, virtual appliance and on-device for Enterprise only; a pricing calculator.Source

Provide free live captions on a screen for ten AI meet-ups a quarter, with a small "captions by Speechmatics" note and a QR code to the free credits. Offer the transcript to organisers afterwards.

Every attendee sees accuracy on real, noisy speech, which is the product's pitch.

• Errors on screen

  Use the best model and good microphones.

• HiringDeveloper Relations Advocate (San Francisco, posted October 22, 2026, $100,000-$140,000): "Be the face of Speechmatics in San Francisco ... bringing back the market insight that shapes our US growth"; "Get developers from first API call to working product"; webinars, hackathons, meet-ups, tutorials and code samples; "Use community and content data to show what's working".Source
• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source

Publish a weekly short: a real accent or mixed-language clip, Speechmatics' live transcript side by side, and a link to try it free. Invite viewers to send their hardest audio.

A recurring format turns one-off views into subscribers and sign-ups.

• Mocking accents

  Feature speakers themselves, with consent.

• Developer ecosystemDeveloper reach: 14 npm packages with about 47,700 weekly downloads (@speechmatics/real-time-client about 18,900); 55 public GitHub repos, the most-starred "speechmatics-academy" (164 stars); a Discord community; docs on Docusaurus; YouTube about 2,220 subscribers, 256 videos and 5.16 million views; LiveKit and Pipecat integrations, and Vapi and Pipecat among the logos on the site.Source
• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source

Offer a grant of credits to one cultural or educational body per language, co-publish results, and build a case study with a broadcaster. Use them on the language pages.

Strong local partners make Speechmatics the default for those languages and generate goodwill and press.

• Small revenue

  Treat as brand and proof.

• Your websiteComparison pages: nine "alternative" pages (AssemblyAI, AWS, Deepgram, ElevenLabs Scribe, Gladia, Google, Microsoft Azure Speech, OpenAI, Soniox) plus a "How we compare" page. Language pages: about 60 /speech-to-text/<language> pages, including Spanish, Portuguese, Welsh, Irish, Basque and Swahili. The site itself is English only.Source
• Your websiteHomepage: "Speech APIs powering Voice AI"; "Multilingual speech-to-text built for regulated industries"; "Real-time speech-to-text in 350ms"; "55+ languages, one model"; HIPAA, GDPR, SOC 2 Type II and ISO 27001; on-device, on-prem or cloud; "We don't log your data as standard". Case studies: AI-Media ("Delivering 120X more"), LiveKit ("Enabling 100,000+ developers"), Adobe (on-device in Premiere), NCI ("99% increase in usage of automated captioning"), Media Track ("20% leap in accuracy"), Prosodica and Stenograph. A banner introduces "Oak, our new medical model built for multilingual consultations".Source