Call transcription (speech-to-text)
Generate text transcripts of your call recordings using a speech-to-text (STT) provider on your own account (BYO model).
When a call has a recording, the CRM can generate a text TRANSCRIPT of the audio using a speech-to-text (STT) provider. The model is BYO ("bring your own account"): you pick the provider (OpenAI Whisper, Deepgram, AssemblyAI or a generic REST endpoint) and enter ITS API key. Transcription runs and is billed on your provider account, with no cost, margin or intermediation from us.
The feature is inert: without an STT configured and enabled, nothing happens, and the call flow stays exactly the same. The API key is treated as a secret: it is encrypted on the server and never shown back.
Prerequisites
- Telephony working with recording (Twilio Voice), since transcription starts from the call recording.
- An account at an STT provider (OpenAI, Deepgram, AssemblyAI) OR your own REST endpoint.
- That provider's API key.
- In Sellio, administrator access to configure the integration.
Supported providers
- OpenAI Whisper: the CRM downloads the recording (using your Twilio Auth Token when the recording is a Twilio URL) and uploads it to transcribe.
- Deepgram: it fetches the recording by URL, which must be reachable by them.
- AssemblyAI: works asynchronously. The CRM sends the URL and follows up until done, and the recording must be reachable by them.
- Generic (REST): any service that accepts a POST with { audio_url } and returns the text. Provide the endpoint.
How to configure
- Open Settings → Integrations and find the "Call transcription (speech-to-text)" card.
- Choose the STT provider.
- Paste the provider API key. Optionally set the model (e.g., whisper-1, nova-2) and, for the generic provider, the endpoint that receives the audio.
- Optional: check "Transcribe automatically when the recording arrives" to transcribe without clicking.
- Click "Save" then "Enable".
How to transcribe a call
- Open a Contact/Lead and go to the "Telephony (calls)" panel.
- On a call with a recording, click "Transcribe".
- Wait: the status shows "Transcribing…" and, when done, the transcript appears and can be expanded.
💡 If you prefer automation, enable "Transcribe automatically": as soon as the recording arrives, the CRM tries to transcribe on its own (best-effort). The "Transcribe" button remains available as a reliable manual path.
How to test
- Make a test call and let it be recorded.
- In the history, confirm the "Play recording" link and click "Transcribe".
- Confirm the transcript appears and reflects the audio.
Troubleshooting
- The "Transcribe" button does not appear: STT is not enabled. Configure and enable the transcription integration.
- Transcription fails (status "Transcription failed"): check the API key and, for Deepgram/AssemblyAI, whether the recording URL is reachable by the provider. With Whisper, the CRM downloads the recording using your Twilio Auth Token.
- Nothing happens to the call flow: correct. Transcription is best-effort and never breaks the call. On failure, only the transcript status is marked failed.
- Cost: in the BYO model, usage goes to YOUR STT provider account; the CRM does not intermediate or charge.