Speechmatics review

Speechmatics sells speech-to-text and voice APIs for teams that need to transcribe live or recorded audio. The homepage says its APIs support more than 55 languages, with natural language switching during a conversation. It lists real-time transcription at 350ms and speech-to-text under one second; those are product claims, not a guarantee for every network or workflow.

The service identifies speakers and includes custom vocabulary. Speechmatics also offers a medical model for dictation and ambient scribing, and says it can cut errors on key terms by up to 50%. For privacy-sensitive deployments, the homepage lists on-device, on-premises and cloud options, alongside HIPAA, GDPR, SOC 2 Type II and ISO 27001 compliance or certification. That mix makes Speechmatics more relevant to multilingual and regulated workflows than a basic transcription app. It is an API product, so teams should also account for integration work and usage-based charges.

The clearest advantage is breadth: one model supports 55+ languages, and the homepage says it can switch languages naturally within a conversation. Speaker diarization and custom vocabulary are included, useful for transcripts where speaker attribution or domain terms matter. Speechmatics also offers a medical model and deployment options ranging from cloud to on-device and on-premises, rather than forcing every workload into a hosted setup.

The free tier gives teams a way to try the API without a monthly plan: the pricing page lists $100 in credits, two concurrent real-time sessions, and up to 10 pre-recorded files per second. Pro rates start at $0.12 per hour for pre-recorded audio and $0.15 per hour for real-time audio. Pro includes 50 concurrent real-time sessions, 10 pre-recorded files per second and unlimited seats. Translation, chapters, topics, summaries and sentiment are available as paid add-ons. One court-reporting Reddit commenter recommends Speechmatics, while another says it works well for them; these are individual reports, not broad evidence of performance.

Predictable monthly budgeting is the main drawback. The pricing page publishes usage-based starting rates, but not Pro subscription prices; it says Pro can be paid as you go, with credit packs or a monthly or annual subscription. Enterprise pricing is also not published. Add-ons can raise the bill: translation is $0.49 per hour where available, while eligible pre-recorded models have separate charges for chapters, topics, summaries and sentiment. PII redaction, entity detection and audio alignment require a price request.

The published self-serve limits matter if volume grows: Free allows two concurrent real-time sessions and 10 pre-recorded files per second; Pro raises real-time concurrency to 50 but keeps the same pre-recorded cap. Enterprise lists concurrency that adjusts to demand and unlimited pre-recorded files per second, but requires a sales quote. Subscription overages use the discounted rate, while other overages use the base rate. A court-reporting Reddit commenter also says paying by the minute may not make sense for every hearing. Speechmatics’ claimed speed and medical error reduction should be validated on your own audio and workflow before committing.

Speechmatics has a free API tier, but paid pricing is usage-based rather than a published monthly plan. The pricing page lists Self-serve Free at $0 with $100 in credits, two concurrent real-time sessions and up to 10 pre-recorded files per second.

Pro starts at $0.12 per audio hour for pre-recorded transcription and $0.15 per hour for real-time. It includes 50 concurrent real-time sessions, 10 pre-recorded files per second and unlimited seats. Pro can be paid as you go, with credit packs, or through a monthly or annual subscription; subscription prices are not published.

Add-ons can raise the bill: translation is $0.49/hour, chapters $0.30/hour, topics $0.15/hour, and summaries and sentiment $0.09/hour each, where eligible. Enterprise SSO is $280/month. Enterprise pricing is quote-only, with concurrency adjusted to demand and unlimited pre-recorded files per second. Watch overages: subscriptions use the discounted rate; otherwise, the base rate applies. The pricing page also lists volume discounts up to 25% per credit and an opt-in 33% discount on Speech-to-Text rates for model training.

Switching to Speechmatics may mean an integration project if you need real-time sessions, custom vocabulary, diarization or a specific deployment. The product material lists APIs and on-device, on-premises and cloud options, but does not document migration tooling or import and export formats. Confirm whether existing transcripts, speaker labels and vocabulary can be carried over before moving production workflows.

The same portability questions apply when leaving: the published sources do not specify export formats, data retention after cancellation or contract lock-in terms. Keep copies of audio and transcripts in a format your next provider accepts, and validate speaker labels and specialized vocabulary in a pilot before cutover. That is a migration precaution, not a documented Speechmatics feature.

We can’t say what tends to make Reddit users leave: no Reddit posts or comments were included in the published material. Cost predictability is a documented consideration, since subscription prices are unpublished and usage is billed per audio hour.