Speechmatics provides speech APIs for voice applications: batch speech-to-text for pre-recorded audio and video, real-time speech-to-text for streaming audio, text-to-speech, and an early-access turn-based Voice Agent API. The same engine can run as a cloud service or on your own infrastructure.
Transcription comes in Standard and Enhanced operating points, plus the Melia 1 model, which handles code-switching between 56 languages within a single real-time stream. Speech intelligence features add structure to the transcript: diarization separates speakers, speaker identification recognizes known individuals, and sentiment, summarization, and topic detection run over the text.
Output can be tailored with custom dictionaries, audio filtering, and audio event tagging, and transcripts can be translated. For self-hosted deployment there are CPU and GPU containers, Kubernetes orchestration, and a virtual appliance.
Features
- Batch and real-time STT: file-based transcription and streaming transcription over the same feature set
- Melia 1: real-time code-switching across 56 languages
- Operating points: Standard and Enhanced accuracy tiers
- Diarization and speaker identification: separate speakers and recognize specific individuals
- Speech intelligence: sentiment analysis, summarization, and topic detection
- Translation: translated output from the transcription stream
- Customization: custom dictionaries, audio filtering, and audio event detection
- Text-to-speech: speech synthesis alongside the recognition APIs
- Voice Agent API: an early-access turn-based API, with LiveKit and Pipecat integrations
- Deployment: SaaS, CPU/GPU containers, Kubernetes, or a virtual appliance
- SDKs and connectors: language SDKs plus Zapier and Vapi integrations
