Speechmatics
Speech-to-text and voice AI APIs you can run in the cloud, on-premises, or on-device
About Speechmatics
Speechmatics is a UK speech recognition company that sells pre-recorded and real-time speech-to-text APIs. Its Melia 1 model transcribes 56 languages with mid-sentence code-switching, Enhanced is its highest-accuracy single-language model, and Linden 1 is a real-time model with built-in turn detection for voice agents. The same models can be deployed in Speechmatics' cloud, on your own servers, or on-device.
Pros & Cons
Pros
- Melia 1 handles mixed-language audio in a single pass at $0.12 per hour, one of the lowest list prices for a multilingual model
- Diarization and word-level timings are included on every model instead of billed as add-ons
- On-prem and on-device deployment options suit regulated industries and offline products
- $100 in free credits with no card is enough to benchmark on your own audio
- Separate medical and voice-agent models let you match the model to the workload
Cons
- Melia 1 is pre-recorded only for now; real-time multilingual code-switching is listed as coming soon
- Free accounts are capped at 2 concurrent real-time sessions
- PII redaction and entity detection are on request rather than self-serve
- Language coverage (55+) is narrower than the 90-100+ languages some competitors list
Key Features
Melia 1 multilingual model
Pre-recorded model that transcribes 56 languages without choosing one up front and switches language mid-sentence. Real-time support is listed as coming soon.
Enhanced model
Highest-accuracy model for single-language audio, with custom dictionary and advanced punctuation, available for both pre-recorded and real-time transcription.
Linden 1 voice agent STT
Real-time-only model with built-in turn detection for conversational voice agents.
Speaker diarization
Speaker labels are included on every model, for pre-recorded and real-time transcription, at no extra charge.
Word-level timings
Every model returns word-level timestamps, along with smart formatting.
Oak 1 medical model
Pre-recorded model tuned for clinical language such as drug names, dosages, and procedure terms.
Add-ons
Translation, chapters, topics, summaries, and sentiment can be added to pre-recorded transcripts for an extra per-hour fee.
Cloud, on-prem, and on-device deployment
The same speech models can run as a SaaS API, inside your own infrastructure, or on-device for privacy-sensitive workloads.
Pricing
- $100 in credits, no card required
- All pre-recorded and real-time models
- 2 concurrent real-time sessions
- Text to speech: first 1 million characters free
- Pre-recorded Melia 1: $0.12/hr
- Pre-recorded Standard: $0.23/hr, Enhanced: $0.38/hr
- Medical Oak 1: $0.15/hr
- Real-time Enhanced: $0.40/hr, Standard: $0.23/hr
- Real-time Linden 1 agent STT: $0.15/hr (listed down from $0.20/hr)
- Diarization and word-level timings included
- 50 concurrent real-time sessions, 10 files per second
- Volume discount up to 25%; model-training discount 33%
- Concurrency that adjusts to demand
- Unlimited files per second
- Dedicated CSM and solutions engineer
- Early feature access
- On-prem and on-device deployment
Best For
Multilingual contact centers
Transcribe calls where agents and customers switch languages mid-conversation without setting a language per call.
Regulated or offline deployments
Run speech recognition on-premises or on-device when audio can't leave your infrastructure.
Voice agents
Use Linden 1 real-time STT with built-in turn detection as the listening layer of a conversational agent.
Clinical documentation
Transcribe clinical conversations and dictation with the Oak 1 medical model.
Similar Tools
Smartly.AI
Generative AI Powered Chatbots for Customer Service
Groq
Ultra-fast AI inference powered by custom LPU silicon
Cohere
Your next breakthrough, powered by AI
D-ID
AI-Powered Video Creation with Talking Avatars
DeepL Write
AI-powered writing assistant for clearer, professional text
EKHOS AI
Secure offline AI transcription for legal, medical, and sensitive recordings