ben.carrasco/jambonz-feature-server
1
0
Fork 0
mirror of https://github.com/jambonz/jambonz-feature-server.git synced 2026-02-13 17:59:42 +00:00
Code Issues Packages Projects Releases Wiki Activity

deepgram: rework continuous asr, and resolve on speech_final not is_f… (#501)

Browse Source 
* deepgram: rework continuous asr, and resolve on speech_final not is_final (wip)

* wip

* deepgram: empty final transcript should trigger resolve with speech if we have buffered transcripts

* wip

* fixes for deepgram compiling multiple transcripts

* test deepgram utteranceEndMs

* more handling of utteranceEndMs

* wip

* better handling of digit strings collected over multiple deepgram responses

* wip

* add support for deepgramOptions.shortUtterance which triggers off of is_final instead of speech_final

* apply deepgram fixes to transcribe

* cleanup continnuous asr

* more continuous asr fixes for deepgram

* update to verb-specifications for handling SttTask properties

* set log level for tests back to error
This commit is contained in:
Dave Horton
2023-10-30 13:57:25 -04:00
committed by GitHub
parent 67f8f7181a
commit f43a5c1491
4 changed files with 179 additions and 60 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
lib/tasks/stt-task.js
View File

@@ -16,12 +16,14 @@ class SttTask extends Task {
normalizeTranscription,
removeSpeechListeners,
setSpeechCredentialsAtRuntime,
compileSonioxTranscripts
compileSonioxTranscripts,
consolidateTranscripts
} = require('../utils/transcription-utils')(logger);
this.setChannelVarsForStt = setChannelVarsForStt;
this.normalizeTranscription = normalizeTranscription;
this.removeSpeechListeners = removeSpeechListeners;
this.compileSonioxTranscripts = compileSonioxTranscripts;
this.consolidateTranscripts = consolidateTranscripts;
this.isHandledByPrimaryProvider = true;
if (this.data.recognizer) {
@@ -138,6 +140,14 @@ class SttTask extends Task {
addKey(key, evt.compiled_context, 3600 * 12)
.catch((err) => this.logger.info({err}, `Error caching cobalt context for ${key}`));
}
_doContinuousAsrWithDeepgram(asrTimeout) {
/* deepgram has an utterance_end_ms property that simplifies things */
assert(this.vendor === 'deepgram');
this.logger.debug(`_doContinuousAsrWithDeepgram - setting utterance_end_ms to ${asrTimeout}`);
const dgOptions = this.data.recognizer.deepgramOptions = this.data.recognizer.deepgramOptions || {};
dgOptions.utteranceEndMs = dgOptions.utteranceEndMs || asrTimeout;
}
}
module.exports = SttTask;