evOo-local/packages/core/src/streaming-voice.test.ts

import { describe, expect, it } from "vitest";
import type { StreamingVoiceFrame } from "./streaming-voice.js";
import {
  ClauseSegmenter,
  calculateRms,
  DEFAULT_ENDPOINT_SILENCE_MS,
  DEFAULT_MIN_SPEECH_FRAMES,
  DEFAULT_SAMPLE_RATE,
  DEFAULT_SILENCE_RMS,
  DEFAULT_VAD_TICK_MS,
  float32ToInt16,
  int16ToFloat32,
  StreamingVad,
} from "./streaming-voice.js";

describe("streaming-voice audio conversions and math", () => {
  it("converts Float32 to Int16 and back accurately", () => {
    const input = new Float32Array([-1.0, -0.5, 0, 0.5, 1.0]);
    const int16 = float32ToInt16(input);
    expect(int16[0]).toBe(-32768);
    expect(int16[2]).toBe(0);
    expect(int16[4]).toBe(32767);

    const backToFloat = int16ToFloat32(int16);
    expect(backToFloat[0]).toBeCloseTo(-1.0, 3);
    expect(backToFloat[1]).toBeCloseTo(-0.5, 3);
    expect(backToFloat[2]).toBeCloseTo(0.0, 3);
    expect(backToFloat[3]).toBeCloseTo(0.5, 3);
    expect(backToFloat[4]).toBeCloseTo(1.0, 3);
  });

  it("clamps Float32 outside [-1, 1]", () => {
    const input = new Float32Array([-2.5, 2.5]);
    const int16 = float32ToInt16(input);
    expect(int16[0]).toBe(-32768);
    expect(int16[1]).toBe(32767);
  });

  it("calculates RMS correctly", () => {
    expect(calculateRms(new Int16Array([]))).toBe(0);

    const silence = new Int16Array(160).fill(0);
    expect(calculateRms(silence)).toBe(0);

    const loud = new Int16Array(100).fill(16384); // ~0.5 amplitude
    expect(calculateRms(loud)).toBeCloseTo(0.5, 2);
  });
});

describe("StreamingVad", () => {
  function makeFrame(amplitude: number, durationMs = DEFAULT_VAD_TICK_MS): StreamingVoiceFrame {
    const samples = Math.floor((DEFAULT_SAMPLE_RATE * durationMs) / 1000);
    const pcm16 = new Int16Array(samples).fill(Math.round(amplitude * 32767));
    return {
      pcm16,
      sampleRate: DEFAULT_SAMPLE_RATE,
      timestamp: Date.now(),
    };
  }

  it("exports correct default constants", () => {
    expect(DEFAULT_ENDPOINT_SILENCE_MS).toBe(120);
    expect(DEFAULT_VAD_TICK_MS).toBe(20);
    expect(DEFAULT_SILENCE_RMS).toBe(0.035);
    expect(DEFAULT_SAMPLE_RATE).toBe(16000);
    expect(DEFAULT_MIN_SPEECH_FRAMES).toBe(3);
  });

  it("suppresses single-frame clicks/pops without triggering speech or endpointing", () => {
    const vad = new StreamingVad();

    // 1 loud frame (click / acoustic pop)
    const popDecision = vad.processFrame(makeFrame(0.2));
    expect(popDecision.isSpeech).toBe(false);
    expect(popDecision.endpointReached).toBe(false);

    // Followed by silence frames
    for (let i = 0; i < 10; i += 1) {
      const silenceDecision = vad.processFrame(makeFrame(0.001));
      expect(silenceDecision.isSpeech).toBe(false);
      // Because speech was never confirmed (needed minSpeechFrames = 3), endpoint never fires
      expect(silenceDecision.endpointReached).toBe(false);
    }
  });

  it("detects sustained speech and triggers endpoint precisely at 120ms (6 ticks)", () => {
    const vad = new StreamingVad();

    // Frame 1: speech energy, but not yet 3 frames
    const f1 = vad.processFrame(makeFrame(0.1));
    expect(f1.isSpeech).toBe(false);
    expect(f1.endpointReached).toBe(false);

    // Frame 2: speech energy
    const f2 = vad.processFrame(makeFrame(0.1));
    expect(f2.isSpeech).toBe(false);
    expect(f2.endpointReached).toBe(false);

    // Frame 3: confirmed speech!
    const f3 = vad.processFrame(makeFrame(0.1));
    expect(f3.isSpeech).toBe(true);
    expect(f3.endpointReached).toBe(false);

    // Frame 4: continued speech
    const f4 = vad.processFrame(makeFrame(0.12));
    expect(f4.isSpeech).toBe(true);
    expect(f4.endpointReached).toBe(false);

    // Now silence begins (each frame is 20ms)
    // Frame 1 of silence: 20ms
    expect(vad.processFrame(makeFrame(0.005)).endpointReached).toBe(false);
    // Frame 2 of silence: 40ms
    expect(vad.processFrame(makeFrame(0.005)).endpointReached).toBe(false);
    // Frame 3 of silence: 60ms
    expect(vad.processFrame(makeFrame(0.005)).endpointReached).toBe(false);
    // Frame 4 of silence: 80ms
    expect(vad.processFrame(makeFrame(0.005)).endpointReached).toBe(false);
    // Frame 5 of silence: 100ms
    expect(vad.processFrame(makeFrame(0.005)).endpointReached).toBe(false);

    // Frame 6 of silence: 120ms -> Endpoint reaches exactly 120ms!
    const endpointDecision = vad.processFrame(makeFrame(0.005));
    expect(endpointDecision.endpointReached).toBe(true);
    expect(endpointDecision.consecutiveSilenceMs).toBe(120);

    // Subsequent silence does not re-fire endpoint
    const postEndpoint = vad.processFrame(makeFrame(0.005));
    expect(postEndpoint.endpointReached).toBe(false);
  });

  it("resets state completely on reset()", () => {
    const vad = new StreamingVad();
    for (let i = 0; i < 5; i += 1) {
      vad.processFrame(makeFrame(0.1));
    }
    vad.reset();

    // After reset, single silence frame has 0 silence duration
    const decision = vad.processFrame(makeFrame(0.001));
    expect(decision.consecutiveSilenceMs).toBe(0);
    expect(decision.endpointReached).toBe(false);
  });
});

describe("ClauseSegmenter", () => {
  it("splits natural clauses on punctuation with minWords threshold", () => {
    const segmenter = new ClauseSegmenter({ minWords: 4, maxWords: 8 });

    // Stream tokens
    const c1 = segmenter.push("Hello there, ");
    // "Hello there," has only 2 words, which is < minWords (4), so no split yet
    expect(c1).toEqual([]);

    const c2 = segmenter.push("I have checked your project, ");
    // Now we have "Hello there, I have checked your project," (7 words) ending in comma!
    expect(c2.length).toBeGreaterThanOrEqual(1);
    expect(c2[0]).toContain("Hello there");

    const c3 = segmenter.push("and everything looks ready to go. ");
    expect(c3.length).toBeGreaterThanOrEqual(1);
    expect(c3[0]).toContain("everything looks ready");
  });

  it("protects abbreviations like Dr. and e.g. from premature splitting", () => {
    const segmenter = new ClauseSegmenter({ minWords: 3 });

    const c1 = segmenter.push("Dr. ");
    expect(c1).toEqual([]);

    const c2 = segmenter.push("Smith arrived at the lab. ");
    expect(c2).toEqual(["Dr. Smith arrived at the lab."]);
  });

  it("protects decimal numbers from premature splitting", () => {
    const segmenter = new ClauseSegmenter({ minWords: 3 });

    const c1 = segmenter.push("The version is 3.14 and ");
    expect(c1).toEqual([]);

    const c2 = segmenter.push("it is stable now. ");
    expect(c2).toEqual(["The version is 3.14 and it is stable now."]);
  });

  it("splits long sentences exceeding maxWords even without punctuation", () => {
    const segmenter = new ClauseSegmenter({ minWords: 3, maxWords: 6 });

    // 8 words without punctuation
    const clauses = segmenter.push("one two three four five six seven eight ");
    expect(clauses.length).toBeGreaterThanOrEqual(1);

    const remaining = segmenter.flush();
    expect(remaining.length).toBeGreaterThanOrEqual(1);
  });

  it("strips markdown code fences using speakable", () => {
    const segmenter = new ClauseSegmenter();

    const clauses = [
      ...segmenter.push("Here is the solution: ```ts\nconsole.log(42);\n```\n It is tested. "),
      ...segmenter.flush(),
    ];
    expect(clauses.some((c) => c.includes("code block") || c.includes("solution"))).toBe(true);
    expect(clauses.some((c) => c.includes("console.log"))).toBe(false);
  });

  it("flushes remaining buffered text on stream completion", () => {
    const segmenter = new ClauseSegmenter({ minWords: 4 });

    const c1 = segmenter.push("Just a short phrase");
    expect(c1).toEqual([]);

    const flushed = segmenter.flush();
    expect(flushed).toEqual(["Just a short phrase"]);
  });

  it("handles reset properly", () => {
    const segmenter = new ClauseSegmenter();
    segmenter.push("Unfinished text here");
    segmenter.reset();
    expect(segmenter.flush()).toEqual([]);
  });
});