Speaker-Separated Transcripts in 100 Languages via Cochl.Sense
Turns speech into a transcript split by speaker turn, with start and end times for each turn. Language is detected automatically across 100 languages. Runs on Cochl's cloud with your own Cochl project key, so it uses 0 CNAPS credits.
🔑 Requires a Cochl project key
What it does
Speech Analysis transcribes a recording and splits it by who is talking. Each entry is one speaker turn: the speaker, what they said, and when.
Speakers are labeled in order of appearance — SPEAKER_00, SPEAKER_01 and so on. If you have registered voice profiles in your Cochl dashboard, matching voices also get that profile's name. Voices that don't match a profile show as "Unknown 0", "Unknown 1" and so on.
There is no language setting. Cochl detects the language on its own.
Problem it solves
Meeting and interview notes | Know who said what, not just what was said |
Podcast and video transcripts | Timed turns per speaker, ready for editing or show notes |
Mixed-language recordings | Automatic language detection across 100 languages |
Input for an LLM | Feed the transcript to a language model for summaries or action items per speaker |
Input/Output
- Input: Sound — one audio file, up to 1 hour. MP3, WAV, FLAC and OGG are sent as-is; other formats are converted first. Starting from a video? Use Video Audio Extract first
- Output: Text — a JSON transcript, one entry per speaker turn:
- speaker — SPEAKER_00, SPEAKER_01, … in order of appearance
- speaker_name — the matching voice profile from your Cochl dashboard, or "Unknown 0", "Unknown 1", … (never empty)
- speaker_score — how well the voice matched a registered profile; -1 when it is unregistered
- transcript — what was said
- start_time / end_time — "MM:SS.ss"
- start_time_sec / end_time_sec — the same times in seconds
- Settings: none
- Requires: A Cochl project key, added under your avatar → Workspace Settings → Integrations
How to Use
- Add your Cochl project key under Workspace Settings → Integrations.
- Upload the recording with Sound Loader. For a video, use Video Loader → Video Audio Extract.
- Connect the audio to Speech Analysis (Cochl) and the output to Text Viewer.
- Run, then check names, numbers and technical terms in the transcript.
Example Workflow
Flow: Sound Loader → Speech Analysis (Cochl) → Text Viewer
Meeting notes: Sound Loader → Speech Analysis (Cochl) → Claude or Gemini ("Summarize this meeting and list action items by speaker") → Text Viewer
Choosing an audio understanding model
You want | Use |
A transcript split by speaker, with timestamps | Speech Analysis (Cochl) |
Plain transcript text, with optional speaker labels | Transcribe (Gemini) |
Timed transcripts on CNAPS servers, no outside key | Speech Recognition |
A quick overview of the whole file — speech, language, setting, events | Audio Insights (Cochl) |
Non-speech sounds with start and end times | Sound Events (Cochl) |
Performance
- Execution: Cochl cloud API; no local GPU
- Length: up to 1 hour of audio per run
- Languages: 100, detected automatically
- Billing: 0 CNAPS credits. Cochl bills your project
- Accuracy: background noise, people talking over each other, accents and specialist vocabulary can affect the transcript and the speaker split
Technical Details
Service | Cochl.Sense Cloud API — Speech Analysis |
Developer | Cochl |
Features | Transcription · speaker separation · optional matching to registered voice profiles · automatic language detection |
Output format | JSON text, one entry per speaker turn with timestamps |
Runtime | Cochl cloud, with your Cochl project key |
Compliance & Provenance
Provider | Cochl |
Provider type | External API |
License / service terms | Cochl service terms and the terms of your Cochl project |
Output provenance | A transcript is extracted speech content; see CNAPS AI Policy for attribution and metadata |
For more on how we classify models and mark outputs, see our AI Policy.
Model Source
- Cochl.Sense
- Cloud API docs: Cochl.Sense Cloud API — Getting started