Tune realtime STT defaults

src/audio/realtime-segmenter.ts

@@ -4,6 +4,13 @@ interface RealtimeSegmenterOptions {
   onSpeechStart?: (peak: number) => void;
   onSpeechDiscarded?: (samples: number) => void;
   onSpeechReady?: (samples: number) => void;
+  preRollSamples?: number;
+  speechStartThreshold?: number;
+  speechContinueThreshold?: number;
+  speechStartFrames?: number;
+  speechEndFrames?: number;
+  minSpeechSamples?: number;
+  maxSpeechSamples?: number;
 }
 
 export class RealtimeSegmenter {
@@ -12,18 +19,27 @@ export class RealtimeSegmenter {
   private readonly speech: number[] = [];
 
   private readonly frameSamples = 320;
-  private readonly preRollSamples = 3200;
-  private readonly speechStartThreshold = 900;
-  private readonly speechContinueThreshold = 450;
-  private readonly speechStartFrames = 2;
-  private readonly speechEndFrames = 18;
-  private readonly minSpeechSamples = 6400;
+  private readonly preRollSamples: number;
+  private readonly speechStartThreshold: number;
+  private readonly speechContinueThreshold: number;
+  private readonly speechStartFrames: number;
+  private readonly speechEndFrames: number;
+  private readonly minSpeechSamples: number;
+  private readonly maxSpeechSamples: number;
 
   private speechActive = false;
   private speechCandidateFrames = 0;
   private silenceFrames = 0;
 
-  constructor(private readonly options: RealtimeSegmenterOptions) {}
+  constructor(private readonly options: RealtimeSegmenterOptions) {
+    this.preRollSamples = options.preRollSamples ?? 3200;
+    this.speechStartThreshold = options.speechStartThreshold ?? 900;
+    this.speechContinueThreshold = options.speechContinueThreshold ?? 450;
+    this.speechStartFrames = options.speechStartFrames ?? 2;
+    this.speechEndFrames = options.speechEndFrames ?? 24;
+    this.minSpeechSamples = options.minSpeechSamples ?? 7200;
+    this.maxSpeechSamples = options.maxSpeechSamples ?? 160000;
+  }
 
   pushChunk(chunk: Buffer): void {
     for (let offset = 0; offset + 1 < chunk.length; offset += 2) {
@@ -78,7 +94,9 @@ export class RealtimeSegmenter {
     }
 
     if (this.silenceFrames < this.speechEndFrames) {
-      return;
+      if (this.speech.length < this.maxSpeechSamples) {
+        return;
+      }
     }
 
     const speechPcm = int16ArrayToBuffer(Int16Array.from(this.speech));