Solaria-1

AI model · Speech to text · Audio / speech → Text by Gladia

Gladia's most multilingual model, optimized for real-time streaming and broad language coverage.

Solaria-1 is Gladia's most versatile speech-to-text model, designed for maximum language coverage and real-time performance. It supports over 100 languages and features native code-switching, allowing it to accurately transcribe conversations where speakers switch languages mid-sentence. The model is the foundation of Gladia's real-time API, delivering partial transcripts in under 103ms and maintaining sub-300ms end-to-end latency for live voice agents and IVR systems. Solaria-1 is trained on real-world noisy audio, ensuring high accuracy on 8kHz telephony streams and variable network conditio

Languages
100+ languages including English, Spani…
Input Modality
Audio, Video
Output Modality
Text (JSON)
Supported Languages
100+
Partial Transcript Latency
<103ms
End-to-End Latency
<300ms
Access Method
WebSocket, REST API, SDK

Overview

Solaria-1 is Gladia's most versatile speech-to-text model, designed for maximum language coverage and real-time performance. It supports over 100 languages and features native code-switching, allowing it to accurately transcribe conversations where speakers switch languages mid-sentence. The model is the foundation of Gladia's real-time API, delivering partial transcripts in under 103ms and maintaining sub-300ms end-to-end latency for live voice agents and IVR systems. Solaria-1 is trained on real-world noisy audio, ensuring high accuracy on 8kHz telephony streams and variable network conditions. It is available for both streaming via WebSockets and asynchronous batch processing.

Multilingual Real-Time Performance

  • 100+ Languages: Supports over 100 languages with automatic detection and native mid-sentence code-switching.
  • 103ms Partial Transcripts: Delivers partial transcripts in under 103ms for near-instant text display during live speech.
  • Sub-300ms Latency: Maintains sub-300ms end-to-end latency, suitable for natural turn-taking in voice agents.
  • Native Code-switching: Handles mid-sentence language switches automatically without requiring manual configuration.

Enterprise Infrastructure

  • Audio Enhancer for Noisy Streams: Features an audio enhancer to clean noisy streams from speakerphones, cars, and patchy networks.
  • Telephony Optimization: Optimized to handle 8kHz telephony audio for contact center and IVR applications.
  • Multi-channel Support: Supports multi-channel transcription with separate streams per participant over a single WebSocket.

Pricing

Real-time rate
0.68€/hr
Async rate (0 - 100 hours)
0.27€/hr

Specifications

Input Modality
Audio, Video
Output Modality
Text (JSON)
Supported Languages
100+
Partial Transcript Latency
<103ms
End-to-End Latency
<300ms
Access Method
WebSocket, REST API, SDK

Added by AntSpace from public sources (gladia.io).

Related

More AI models by Gladia