Intermediate results on using session_client.streaming_detect_intent() about python-dialogflow

[eniasebiomo]

Having same issue

[dysnomia]

I noticed that if you don't group the audio chunks in a iterator collection (yield), they will not be handled as an unique stream when calling streaming_detect_intent(), so you need to process the stream until you somehow detects the audio input stopped or something similar and then deliver the iterator collection for detection, which is kind of silly since we could have detected intents prior to the end of the streaming (almost the same effect of recording the audio and then calling detect_intent()).

The generator you pass into sessions_client.streaming_detect_intent and the partial Speech recognition responses you get are processed concurrently. That is, you can make the generator dependent on the partial recognition results (http://dialogflow-python-client-v2.readthedocs.io/en/latest/gapic/v2/types.html#dialogflow_v2.types.StreamingDetectIntentResponse).

Stopping: In particular, you decide when to stop Speech recognition by exiting the generator based on whatever signals you want. If you set single_utterance to True (http://dialogflow-python-client-v2.readthedocs.io/en/latest/gapic/v2/types.html#dialogflow_v2.types.StreamingDetectIntentRequest.single_utterance), we will return a RECOGNITION_EVENT_END_OF_SINGLE_UTTERANCE event when we believe a Speech utterance has ended. You should then exit the generator (in fact we ignore further input audio).

Warning: As described in http://dialogflow-python-client-v2.readthedocs.io/en/latest/gapic/v2/types.html#dialogflow_v2.types.StreamingRecognitionResult you will get exactly one intent result per stream (returning multiple intents wouldn't have an obvious semantic). The main use of StreamingDetectIntent is to send audio that is being recorded from a microphone, get partial recognition results and then close the microphone after RECOGNITION_EVENT_END_OF_SINGLE_UTTERANCE. There is no significant advantage in using StreamingDetectIntent if you have prerecorded audio.

Is there a way to call it separately and inform it is the same stream for getting intermediate results, thus dynamically decide to stop streaming and process the results?

No, separate streams are fully independent of each other.

Please let me know whether that clarified the use of StreamingDetectIntent.

[nnegrey]

This appears to be answered, if not, please feel free to re-open.

Thanks! :)