Critical Issue: Speech-to-Text (STT) Unusable on Voiceflow Phone/Call Test Channel
❓Voice
Hello everyone,
I'm facing a critical problem with my Voiceflow voice agent that I'm developing for a pizza ordering system. Despite my efforts, the Speech-to-Text (STT) is not working correctly: it transcribes random, incoherent words, even when I speak clearly.
I've managed to isolate the failure, but I can't solve it.
Project Context
Orchestration Platform: Voiceflow (Voice Project).
Project Language: French (FR).
Problematic Channel: The internal Voiceflow phone call simulation / Twilio Channel.
Diagnostic Results (Key Findings)
Audio Quality (Eliminated): I checked the recording of my voice within the Voiceflow interface. The recording is perfectly clear and of good quality. The audio is therefore reaching the platform correctly.
Browser Test (OK): When I use the default testing mode (direct microphone via browser), voice recognition works correctly (Wideband ASR).
"Call" Test (TOTAL FAILURE): As soon as I switch to the "Call" simulation mode, the STT completely breaks down.
Voice input Engines Tested
I verified the voice input Language setting to French (FR) and tested with the following engines:
- Deepgram
- Cartesia
The result is the same: total failure in call mode.
I'm facing a critical problem with my Voiceflow voice agent that I'm developing for a pizza ordering system. Despite my efforts, the Speech-to-Text (STT) is not working correctly: it transcribes random, incoherent words, even when I speak clearly.
I've managed to isolate the failure, but I can't solve it.
Project Context
Orchestration Platform: Voiceflow (Voice Project).
Project Language: French (FR).
Problematic Channel: The internal Voiceflow phone call simulation / Twilio Channel.
Diagnostic Results (Key Findings)
Audio Quality (Eliminated): I checked the recording of my voice within the Voiceflow interface. The recording is perfectly clear and of good quality. The audio is therefore reaching the platform correctly.
Browser Test (OK): When I use the default testing mode (direct microphone via browser), voice recognition works correctly (Wideband ASR).
"Call" Test (TOTAL FAILURE): As soon as I switch to the "Call" simulation mode, the STT completely breaks down.
Voice input Engines Tested
I verified the voice input Language setting to French (FR) and tested with the following engines:
- Deepgram
- Cartesia
The result is the same: total failure in call mode.
