I'm encountering an STT issue where caller audio is detected in recordings, but no transcript is generated, so the assistant stays silent after the first message. Symptom: Caller audio is present in the recording (caller repeatedly says “hello”), but no user transcript is produced, so the assistant never responds. Transcriber configs tested: Google STT → no transcript produced ElevenLabs (model: scribe_v2) → same issue Deepgram Nova 2 → works but transcription quality is poor Context: Our calls require Chinese + English transcription, and Deepgram struggles with this mixed language. Has anyone experienced audio being detected but STT not returning transcripts with Google or ElevenLabs? Could this be a configuration issue or pipeline problem? Any help would be greatly appreciated 🙏

Suggestion: Set up a custom transcriber to gain full control and debug your STT pipeline for Chinese + English calls. Start with the following Node.js project template:

bash
mkdir vapi-custom-transcriber
cd vapi-custom-transcriber
npm init -y
npm install ws express dotenv @deepgram/sdk

This lets you forward audio to your preferred STT provider and log/debug transcript events. See the [Vapi custom transcriber guide](https://docs.vapi.ai/customization/custom-transcriber#project-setup) for full implementation details. Source: - [Vapi Custom Transcriber Documentation](https://docs.vapi.ai/customization/custom-transcriber#project-setup)

It sounds like the audio is reaching the recording pipeline but the STT layer isn’t emitting transcripts back to the assistant, which is why the conversation stalls after the greeting. I’ve run into similar situations before where the issue ended up being a config or streaming mismatch rather than the model itself, especially with mixed-language audio. If you’d like, share a bit more about how your STT pipeline is wired and I can take a quick look happy to help you narrow it down @zie

Hi, could you please provide us with the affected call IDs?