带有音频的 Dialogflow V2 HTTP API detectIntent 返回无效响应（不完整的正文）

Question

I faced with next issue: some of Dialogflow V2 HTTP API requests response with incomplete body. No regularity found, it can happen during random requests with different intent & session info.

Example:

POST https://dialogflow.googleapis.com/v2/projects/{project-name}/agent/sessions/{session-uuid}:detectIntent

> Content-Type: application/json
> Authorization: Bearer {token}

REQUEST BODY:
{
    "queryInput": {
        "audioConfig": {
            "audioEncoding": "AUDIO_ENCODING_OGG_OPUS",
            "sampleRateHertz": 48000,
            "languageCode": "en-US",
            "model": "command_and_search"
        }
    },
    "inputAudio": "{base64-encoded-file}"
}

RESPONSE:
< HTTP/1.1 200 OK
< Content-Type: application/json; charset=UTF-8

RESPONSE BODY:
{
  "queryResult": {
    "languageCode": "en-US"
  }
}

... and that's all. API returns body with languageCode field only in queryResult . All are missing except this one.

Issue happen only during requests with audio, all works fine for text input.

Any help/tips how to avoid this issue? I would be grateful for any help.

Answer 1

I have checked out the documentation [1] and it seems to me you are using a sampling interval different than the sampling interval of the default format.

Particularly, as you can see "Opus encoded audio frames in Ogg container (OggOpus). sampleRateHertz must be 16000."

Meanwhile you specified "sampleRateHertz": 48000.

This may be the reason of your unexpected results. If the sampling rate are different, then you may have a wrong identification of the signal. I would suggest you to resample the audio in input to 16000 Hz or to change the encoding format, and perhaps opting for FLAC (Free Lossless Audio Codec) because is the recommended encoding because it is lossless (therefore recognition is not compromised)

[1] https://cloud.google.com/dialogflow/docs/reference/rest/v2beta1/QueryInput#audioencoding