Speech-to-text
Full-context transcription at linear cost: re-decoding’s coherence at chunking’s price, any turn length.
The engine’s own streaming speech-to-text, the fourth stage of the pipeline — and the first model available on its own. Speech-to-text ships first.
Accuracy vs. speed
Measured, realtime streaming. Every point measured by us on identical samples: avg WER across both test sets (x) vs p50 finalization latency, end of speech → final transcript (y). Closer to the bottom-left is better.
| System | Avg. word error rate (measured) | Final latency p50 (ms) |
|---|---|---|
| OpenAI gpt-4o-transcribe realtime | ≈14% | ≈715 ms |
| Qwen3-ASR naive 300ms re-decode | ≈8.5% | ≈290 ms |
| AssemblyAI streaming | ≈8.9% | ≈125 ms |
| Modulate vfast streaming | ≈17.2% | ≈120 ms |
| ElevenLabs Realtime | ≈8.3% | ≈100 ms |
| Deepgram Nova-3 | ≈13% | ≈100 ms |
| Alebex ASR | ≈8.5% | ≈40 ms |
Values are read off the plotted positions against the chart’s own axes; the measurement publishes positions rather than a table. Measured by Alebex on identical samples across both test sets.
Context-aware linear streaming speech-to-text
Re-decoding keeps the whole turn in context and pays for it again at every step. Chunking pays a flat price and throws the context away. This keeps the context and the flat price.
| Approach | How it is arranged | What each step sees | What each step costs |
|---|---|---|---|
| Re-decoding | Decode the whole turn again at every step | The whole turn so far | Grows with the turn |
| Chunking | Decode each chunk on its own | The current chunk only | Flat, and the context is gone |
| Alebex | Carry the context, pay the chunk price | The whole turn so far | Flat, any turn length |
How three approaches trade context against cost across one turn. Arrangement, not to scale. The blocks show whether the work grows with the turn, not how long it takes.
Open a session, stream the audio, read the transcript
REST creates and manages the session. The WebSocket carries the audio for as long as the session is live.
Shell request with cURL. Illustrative sample. Endpoint paths, parameter names and field names are confirmed in the documentation issued with your account.
Illustrative sample of the message sequence, not a recording of a measured run. No timing is shown, and neither transport is presented as better than the other.
Try the live demo
Drop in your website. Alebex reads it, learns what you do, and calls you in under 30 seconds — fully trained, zero setup.