Gladia is a speech-to-text and audio intelligence API for building voice products. It offers two transcription modes: an asynchronous endpoint for recordings and long-form audio or video, and a real-time WebSocket endpoint for live voice applications with sub-300 ms latency.
The underlying models are the Solaria family. Solaria-1 is positioned as a multilingual model covering a broad set of languages, and Solaria-3 improves accuracy on real English audio and on French, German, Spanish, and Italian, reporting 9.6% word error rate on English recordings.
On top of transcription, an audio intelligence layer enriches transcripts with LLM-powered analysis without calling a separate service. It plugs into the common voice-agent and telephony stacks and into no-code automation tools, and a desktop dictation app, GladiaFlow, uses the same real-time engine.
Features
- Async transcription: REST endpoint for pre-recorded audio and video files
- Real-time transcription: WebSocket streaming with sub-300 ms latency
- Solaria models: Solaria-1 for broad multilingual coverage, Solaria-3 for English and major European languages
- Diarization and language handling: speaker separation, automatic language detection, and code switching
- Translation: translated transcripts alongside the source language
- Audio intelligence: summarisation, chapterisation, sentiment and emotion analysis, named entity recognition, and PII redaction
- Custom vocabulary and spelling: domain terms and preferred spellings applied at transcription time
- Integrations: Pipecat, LiveKit, Vapi, Recall, Twilio, VideoSDK, Composio, Zapier, Make, and n8n
- Developer tooling: official SDK, API reference, and an interactive playground