I’m prototyping a custom voice biometrics integration. In our Architect inbound flow, I’m using the ‘Record Audio’ action to capture a 5-second sample of the caller speaking a passphrase. I then pass that recording to an external AWS Lambda via a Data Action for biometric authentication.
However, the 3rd party API is consistently rejecting the audio file due to low fidelity. I pulled the temporary file from the API payload and found that the Edge is compressing the audio snippet extremely aggressively (8kHz, very low bitrate). Is there a way to configure the ‘Record Audio’ block in Architect to capture uncompressed PCM WAV or at least a higher-quality G.711 stream specifically for biometric analysis?
The ‘Record Audio’ block in Architect is primarily designed for voicemail drops and simple prompts, not DSP-level analysis. It defaults to highly compressed formats to save storage space on the platform. You cannot override the codec or bitrate within the Architect UI for that specific action block.
If you need high-fidelity audio for biometrics, you shouldn’t use the Architect recording block. Instead, you should use the ‘AudioHook’ (formerly Media Helper) integration. AudioHook allows you to stream the live, uncompressed RTP media directly from the Edge to your third-party biometric engine in real-time. It’s built exactly for this use case.
I’ve dealt with this from the Salesforce side where we tried to build our own emotion detection.
The point above is correct. If you absolutely have to use a static file instead of an AudioHook stream, you have to intercept it at the Trunk level. We ended up using a ‘Dual-Channel’ recording policy that saves the call to the S3 bucket in higher quality, and then our external system pulls the recording via the API after the authentication segment. But for real-time authentication in the IVR, AudioHook is the only supported path to bypass Architect’s compression.