add support for getting partial transcripts when doing background gather

gather: do not set utterance_end_ms on Deepgram if endpointing is explicitly turned off, and also in that case return transcripts when is_final is true
2026-02-09 02:30:17 +00:00 · 2024-04-17 16:32:41 -04:00 · 2024-04-17 14:10:10 -04:00
5 changed files with 12 additions and 11 deletions
--- a/lib/tasks/config.js
+++ b/lib/tasks/config.js
@@ -30,7 +30,7 @@ class TaskConfig extends Task {
        input: ['speech']
      };
      [
-        'finishOnKey', 'input', 'numDigits', 'minDigits', 'maxDigits',
+        'finishOnKey', 'input', 'numDigits', 'minDigits', 'maxDigits', 'partialResultHook',
        'interDigitTimeout', 'bargein', 'dtmfBargein', 'minBargeinWordCount', 'actionHook'
      ].forEach((k) => {
        if (this.bargeIn[k]) this.gatherOpts[k] = this.bargeIn[k];
--- a/lib/tasks/gather.js
+++ b/lib/tasks/gather.js
@@ -351,7 +351,8 @@ class TaskGather extends SttTask {
    /* some special deepgram logic */
    if (this.vendor === 'deepgram') {
      if (this.isContinuousAsr) this._doContinuousAsrWithDeepgram(this.asrTimeout);
-      if (this.data.recognizer?.deepgramOptions?.shortUtterance) this.shortUtterance = true;
+      if (this.data.recognizer?.deepgramOptions?.shortUtterance ||
+        this.data.recognizer?.deepgramOptions?.endpointing === false) this.shortUtterance = true;
    }

    const opts = this.setChannelVarsForStt(this, this.sttCredentials, this.language, this.data.recognizer);
--- a/lib/utils/transcription-utils.js
+++ b/lib/utils/transcription-utils.js
@@ -692,7 +692,7 @@ module.exports = (logger) => {
        ...('endpointing' in deepgramOptions) &&
          {DEEPGRAM_SPEECH_ENDPOINTING: deepgramOptions.endpointing === false ? 'false' : deepgramOptions.endpointing,
            // default DEEPGRAM_SPEECH_UTTERANCE_END_MS is 1000, will be override by user settings later if there is.
-            DEEPGRAM_SPEECH_UTTERANCE_END_MS: 1000},
+            DEEPGRAM_SPEECH_UTTERANCE_END_MS: deepgramOptions.endpointing === false ? null : 1000},
        ...(deepgramOptions.utteranceEndMs) &&
          {DEEPGRAM_SPEECH_UTTERANCE_END_MS: deepgramOptions.utteranceEndMs},
        ...(deepgramOptions.vadTurnoff) &&
--- a/package-lock.json
+++ b/package-lock.json
@@ -18,7 +18,7 @@
        "@jambonz/speech-utils": "^0.0.51",
        "@jambonz/stats-collector": "^0.1.9",
        "@jambonz/time-series": "^0.2.8",
-        "@jambonz/verb-specifications": "^0.0.69",
+        "@jambonz/verb-specifications": "^0.0.70",
        "@opentelemetry/api": "^1.8.0",
        "@opentelemetry/exporter-jaeger": "^1.23.0",
        "@opentelemetry/exporter-trace-otlp-http": "^0.50.0",
@@ -2360,9 +2360,9 @@
      }
    },
    "node_modules/@jambonz/verb-specifications": {
-      "version": "0.0.69",
-      "resolved": "https://registry.npmjs.org/@jambonz/verb-specifications/-/verb-specifications-0.0.69.tgz",
-      "integrity": "sha512-DWnz7XRkCzpzyCVJH7NtScv+wSlUC414/EO8j/gPZs3RT4WBW1OBXwXpfjURHcSrDG7lycz+tfA+2WoUdW/W+g==",
+      "version": "0.0.70",
+      "resolved": "https://registry.npmjs.org/@jambonz/verb-specifications/-/verb-specifications-0.0.70.tgz",
+      "integrity": "sha512-WRSM7+hMnxLv4Fe5PG02m64uKX1AryRLF/yWpv/0PjHJSO4SYXN8CCAhgb1vctq27FMenMtOrSVt0haZeARObg==",
      "dependencies": {
        "debug": "^4.3.4",
        "pino": "^8.8.0"
@@ -11968,9 +11968,9 @@
      }
    },
    "@jambonz/verb-specifications": {
-      "version": "0.0.69",
-      "resolved": "https://registry.npmjs.org/@jambonz/verb-specifications/-/verb-specifications-0.0.69.tgz",
-      "integrity": "sha512-DWnz7XRkCzpzyCVJH7NtScv+wSlUC414/EO8j/gPZs3RT4WBW1OBXwXpfjURHcSrDG7lycz+tfA+2WoUdW/W+g==",
+      "version": "0.0.70",
+      "resolved": "https://registry.npmjs.org/@jambonz/verb-specifications/-/verb-specifications-0.0.70.tgz",
+      "integrity": "sha512-WRSM7+hMnxLv4Fe5PG02m64uKX1AryRLF/yWpv/0PjHJSO4SYXN8CCAhgb1vctq27FMenMtOrSVt0haZeARObg==",
      "requires": {
        "debug": "^4.3.4",
        "pino": "^8.8.0"
--- a/package.json
+++ b/package.json
@@ -34,7 +34,7 @@
    "@jambonz/speech-utils": "^0.0.51",
    "@jambonz/stats-collector": "^0.1.9",
    "@jambonz/time-series": "^0.2.8",
-    "@jambonz/verb-specifications": "^0.0.69",
+    "@jambonz/verb-specifications": "^0.0.70",
    "@opentelemetry/api": "^1.8.0",
    "@opentelemetry/exporter-jaeger": "^1.23.0",
    "@opentelemetry/exporter-trace-otlp-http": "^0.50.0",
Author	SHA1	Message	Date
Dave Horton	d4962c2893	add support for getting partial transcripts when doing background gather	2024-04-17 16:32:41 -04:00
Dave Horton	cd06fa2049	gather: do not set utterance_end_ms on Deepgram if endpointing is explicitly turned off, and also in that case return transcripts when is_final is true	2024-04-17 14:10:10 -04:00