Speechmatics

Converts live audio streams to text with claimed sub-1-second latency using a streaming API architecture that processes audio chunks incrementally rather than waiting for complete audio files. The system maintains persistent connections for continuous audio input and outputs partial/final transcription results as they become available, enabling real-time voice agent applications and live captioning use cases.

Processes pre-recorded audio files asynchronously, transcribing them into text across 55+ languages and dialects using a job-based queue system. Files are submitted to a batch processing pipeline that handles transcription at a rate of up to 10 jobs per second (Pro tier), returning complete transcripts with speaker identification and confidence metadata once processing completes.

Offers a startup program providing up to $50,000 in API credits for eligible early-stage companies, reducing the cost of speech recognition for bootstrapped teams and accelerating adoption in startups. Credits can be applied to both speech-to-text and text-to-speech usage, enabling startups to build voice-enabled products without significant upfront infrastructure costs.

Provides native integration with LiveKit, an open-source voice agent framework, enabling developers to build real-time voice agents using Speechmatics speech recognition and synthesis. The integration handles audio streaming, transcription, and response generation within the LiveKit agent architecture, simplifying the development of conversational AI applications.

Provides a free tier allowing developers to test speech recognition and synthesis capabilities with 480 minutes of monthly transcription and 1 million characters of monthly text-to-speech synthesis. The free tier includes access to real-time and batch transcription across all 55+ languages, enabling developers to prototype voice applications without upfront costs.

Provides a paid tier at $0.24 per hour of transcription with a 20% discount available for volume commitments. The Pro tier includes 480 minutes of free monthly transcription (matching free tier) plus overage billing, 50 concurrent sessions for real-time transcription, and 10 file jobs per second for batch processing. Pricing structure and overage rates are not fully documented.

Allows users to define custom words, phrases, and domain-specific terminology that the speech recognition model should prioritize during transcription. Custom dictionaries are injected into the transcription pipeline to improve accuracy for specialized vocabulary (medical terms, product names, technical jargon) that may not be well-represented in the base model's training data.

Automatically identifies and segments audio by speaker, labeling different speakers in transcripts and providing speaker-aware transcription output. The system uses speaker diarization algorithms to detect speaker boundaries and assign consistent speaker identities throughout the audio, enabling multi-party conversation transcription without manual speaker labeling.

Provides a specialized speech recognition model trained on medical terminology and healthcare-specific language, achieving 50% error reduction on key medical terms compared to the base model. The Medical Model is optimized for transcribing clinical notes, patient interviews, and healthcare conversations where accurate terminology is critical for compliance and patient safety.

Converts text to natural-sounding speech with optimized latency for real-time voice agent applications. The TTS engine is designed to minimize synthesis delay, enabling voice agents to respond quickly to user input without noticeable pauses. Currently supports English with additional languages coming soon.

Provides deployment options beyond cloud SaaS, including on-premises installation and on-device execution (demonstrated with Adobe Premiere integration). This enables organizations with strict data privacy requirements to run speech recognition locally without sending audio to cloud servers. The on-device model is optimized to run efficiently on standard hardware (laptops, edge devices) while maintaining accuracy comparable to cloud models.

Provides cloud-based deployment across multiple geographic regions with configurable redundancy and failover for enterprise customers. Enterprise tier customers can select deployment regions and configure high-availability setups to meet geographic data residency requirements and ensure service continuity across regions.

Provides end-to-end encryption for audio data in transit and at rest, with encryption key management controlled by the customer. Enterprise customers can configure encryption policies to ensure sensitive audio content (medical records, legal proceedings, confidential business calls) is protected throughout the transcription pipeline.

Extracts precise timing information for each word in the transcript, enabling synchronization of transcription with video, animation, or other time-based media. The system provides word-level timestamps and confidence scores, allowing developers to build features like auto-generated captions, lip-sync animation, and searchable transcript navigation.