refactor(audio,stage-ui): simplify code

packages/audio/package.json

@@ -42,7 +42,8 @@
   "scripts": {
     "dev": "pnpm run build",
     "build": "tsdown",
-    "typecheck": "tsc --noEmit"
+    "typecheck": "tsc --noEmit",
+    "test:run": "vitest run"
   },
   "peerDependencies": {
     "vue": ">=3"

packages/audio/src/encoding/wav.test.ts

@@ -0,0 +1,33 @@
+import { describe, expect, it } from 'vitest'
+
+import { toWav, toWavFromPCM16 } from './wav'
+
+describe('toWav', () => {
+  it('converts Float32 samples to PCM16 bytes by default', () => {
+    const samples = new Float32Array([-1, 0, 1])
+    const wav = toWav(samples.buffer, 24000)
+    const view = new DataView(wav)
+
+    expect(view.getInt16(44, true)).toBe(-32768)
+    expect(view.getInt16(46, true)).toBe(0)
+    expect(view.getInt16(48, true)).toBe(32767)
+  })
+
+  it('preserves PCM16 bytes and writes the WAV metadata', () => {
+    const pcmBytes = new Uint8Array([0x00, 0x80, 0xFF, 0x7F])
+    const wav = toWavFromPCM16(pcmBytes, 24000)
+    const view = new DataView(wav)
+
+    expect(new TextDecoder().decode(new Uint8Array(wav, 0, 4))).toBe('RIFF')
+    expect(view.getUint32(4, true)).toBe(40)
+    expect(new TextDecoder().decode(new Uint8Array(wav, 8, 4))).toBe('WAVE')
+    expect(view.getUint16(20, true)).toBe(1)
+    expect(view.getUint16(22, true)).toBe(1)
+    expect(view.getUint32(24, true)).toBe(24000)
+    expect(view.getUint32(28, true)).toBe(48000)
+    expect(view.getUint16(32, true)).toBe(2)
+    expect(view.getUint16(34, true)).toBe(16)
+    expect(view.getUint32(40, true)).toBe(pcmBytes.byteLength)
+    expect([...new Uint8Array(wav, 44)]).toEqual([...pcmBytes])
+  })
+})

packages/audio/src/encoding/wav.ts

@@ -6,46 +6,66 @@ function writeString(dataView: DataView, offset: number, string: string) {
   }
 }
 
-export function toWav(buffer: ArrayBufferLike, sampleRate: number, channel = 1) {
-  const samples = new Float32Array(buffer) // allows indexing
-  const numChannels = channel
-  const numSamples = samples.length
-
-  // Create the WAV file container
-  const arrayBuffer = new ArrayBuffer(44 + numSamples * 2)
+function createWavBuffer(dataSize: number, sampleRate: number, channel: number): ArrayBuffer {
+  const bitsPerSample = 16
+  const byteRate = sampleRate * channel * (bitsPerSample / 8)
+  const blockAlign = channel * (bitsPerSample / 8)
+  const arrayBuffer = new ArrayBuffer(44 + dataSize)
   const dataView = new DataView(arrayBuffer)
 
-  // RIFF chunk descriptor
   writeString(dataView, 0, 'RIFF')
-  dataView.setUint32(4, 36 + numSamples * 2, true)
+  dataView.setUint32(4, 36 + dataSize, true)
   writeString(dataView, 8, 'WAVE')
 
-  // fmt sub-chunk
   writeString(dataView, 12, 'fmt ')
   dataView.setUint32(16, 16, true)
-  dataView.setUint16(20, 1, true) // PCM format
-  dataView.setUint16(22, numChannels, true)
+  dataView.setUint16(20, 1, true)
+  dataView.setUint16(22, channel, true)
   dataView.setUint32(24, sampleRate, true)
-  dataView.setUint32(28, sampleRate * numChannels * 2, true) // byte rate
-  dataView.setUint16(32, numChannels * 2, true) // block align
+  dataView.setUint32(28, byteRate, true)
+  dataView.setUint16(32, blockAlign, true)
+  dataView.setUint16(34, bitsPerSample, true)
 
-  dataView.setUint16(34, 16, true) // bits per sample
-
-  // data sub-chunk
   writeString(dataView, 36, 'data')
-  dataView.setUint32(40, numSamples * 2, true)
+  dataView.setUint32(40, dataSize, true)
 
-  // PCM samples
-  const offset = 44
-  for (let i = 0; i < numSamples; i++) {
+  return arrayBuffer
+}
+
+/**
+ * Encodes Float32 samples as a WAV file.
+ *
+ * @example
+ * toWav(float32Samples.buffer, 24000)
+ * // => WAV data with converted PCM16 samples
+ */
+export function toWav(buffer: ArrayBufferLike, sampleRate: number, channel = 1): ArrayBuffer {
+  const samples = new Float32Array(buffer)
+  const arrayBuffer = createWavBuffer(samples.length * 2, sampleRate, channel)
+  const dataView = new DataView(arrayBuffer)
+
+  for (let i = 0; i < samples.length; i++) {
     const sample = Math.max(-1, Math.min(1, samples[i]))
     const value = sample < 0 ? sample * 0x8000 : sample * 0x7FFF
-    dataView.setInt16(offset + i * 2, value, true)
+    dataView.setInt16(44 + i * 2, value, true)
   }
 
   return arrayBuffer
 }
 
+/**
+ * Wraps raw signed 16-bit PCM samples in a WAV file.
+ *
+ * @example
+ * toWavFromPCM16(pcmBytes, 24000)
+ * // => WAV data with the original PCM16 bytes
+ */
+export function toWavFromPCM16(pcmBytes: Uint8Array, sampleRate: number, channel = 1): ArrayBuffer {
+  const arrayBuffer = createWavBuffer(pcmBytes.byteLength, sampleRate, channel)
+  new Uint8Array(arrayBuffer, 44).set(pcmBytes)
+  return arrayBuffer
+}
+
 export function toWAVBase64(buffer: ArrayBufferLike, sampleRate: number) {
   return encodeBase64(toWav(buffer, sampleRate))
 }

packages/audio/vitest.config.ts

@@ -0,0 +1,7 @@
+import { defineConfig } from 'vitest/config'
+
+export default defineConfig({
+  test: {
+    include: ['src/**/*.test.ts'],
+  },
+})

packages/stage-ui/src/stores/providers/google-gemini-speech.ts

@@ -2,6 +2,8 @@ import type { SpeechProviderWithExtraOptions } from '@xsai-ext/providers/utils'
 
 import type { ModelInfo, ProviderMetadata, VoiceInfo } from '../providers'
 
+import { toWavFromPCM16 } from '@proj-airi/audio/encoding'
+
 const PROVIDER_ID = 'google-gemini-audio-speech'
 const DEFAULT_BASE_URL = 'https://generativelanguage.googleapis.com/v1beta'
 const DEFAULT_MODEL = 'gemini-2.5-flash-preview-tts'
@@ -45,42 +47,6 @@ const GOOGLE_GEMINI_TTS_VOICES: [string, string][] = [
   ['Sulafat', 'Warm'],
 ]
 
-/** Wraps raw PCM16 mono data in a minimal WAV container. */
-function wrapPCM16InWAV(pcmBytes: Uint8Array, sampleRate = 24000): Uint8Array {
-  const numChannels = 1
-  const bitsPerSample = 16
-  const byteRate = sampleRate * numChannels * (bitsPerSample / 8)
-  const blockAlign = numChannels * (bitsPerSample / 8)
-  const header = new ArrayBuffer(44)
-  const view = new DataView(header)
-
-  const writeStr = (offset: number, str: string) => {
-    for (let i = 0; i < str.length; i++)
-      view.setUint8(offset + i, str.charCodeAt(i))
-  }
-
-  writeStr(0, 'RIFF')
-  view.setUint32(4, 36 + pcmBytes.length, true)
-  writeStr(8, 'WAVE')
-
-  writeStr(12, 'fmt ')
-  view.setUint32(16, 16, true)
-  view.setUint16(20, 1, true)
-  view.setUint16(22, numChannels, true)
-  view.setUint32(24, sampleRate, true)
-  view.setUint32(28, byteRate, true)
-  view.setUint16(32, blockAlign, true)
-  view.setUint16(34, bitsPerSample, true)
-
-  writeStr(36, 'data')
-  view.setUint32(40, pcmBytes.length, true)
-
-  const wav = new Uint8Array(44 + pcmBytes.length)
-  wav.set(new Uint8Array(header), 0)
-  wav.set(pcmBytes, 44)
-  return wav
-}
-
 /** Decodes a base64 string into a Uint8Array. */
 function base64ToBytes(base64: string): Uint8Array {
   const binaryString = atob(base64)
@@ -169,12 +135,9 @@ function createAudioFetch(apiKey: string, baseUrl: string) {
     }
 
     const pcmBytes = base64ToBytes(audioBase64)
-    const wavBytes = wrapPCM16InWAV(pcmBytes)
+    const wavBuffer = toWavFromPCM16(pcmBytes, 24000)
 
-    // NOTICE: wrapPCM16InWAV always creates a fresh Uint8Array, so .buffer is the full
-    // backing ArrayBuffer (not a subarray view into a larger buffer). The `as ArrayBuffer`
-    // cast is needed because .buffer returns ArrayBufferLike in newer TypeScript.
-    return new Response(wavBytes.buffer as ArrayBuffer, {
+    return new Response(wavBuffer, {
       status: 200,
       headers: { 'Content-Type': 'audio/wav' },
     })

packages/stage-ui/src/stores/providers/openai-compatible-builder.ts

@@ -20,21 +20,6 @@ function normalizeBaseUrl(value: unknown): string {
   return base
 }
 
-function shouldLog(): boolean {
-  try {
-    // Opt-in via localStorage to minimize I/O in production
-    return typeof localStorage !== 'undefined' && localStorage.getItem('airi:debug') === '1'
-  }
-  catch {
-    return false
-  }
-}
-
-function logWarn(...args: unknown[]) {
-  if (shouldLog())
-    console.warn(...args)
-}
-
 /**
  * Wraps transcription providers so OpenAI audio options like `language` and `prompt` are preserved.
  */
@@ -194,8 +179,8 @@ export function buildOpenAICompatibleProvider(
             detected = models[0].id
         }
         catch (e) {
-          logWarn(`Model auto-detection failed: ${(e as Error).message}`)
-          logWarn('Falling back to default test model for validation checks.')
+          console.warn(`Model auto-detection failed: ${(e as Error).message}`)
+          console.warn('Falling back to default test model for validation checks.')
           try {
             if (capabilities?.listModels) {
               const models = await capabilities.listModels(config)
@@ -206,7 +191,7 @@ export function buildOpenAICompatibleProvider(
             }
           }
           catch (e) {
-            logWarn(`Model auto-detection via capabilities.listModels also failed: ${(e as Error).message}`)
+            console.warn(`Model auto-detection via capabilities.listModels also failed: ${(e as Error).message}`)
           }
         }
         return detected

packages/stage-ui/src/stores/providers/openrouter/audio-speech.ts

@@ -2,6 +2,8 @@ import type { SpeechProvider } from '@xsai-ext/providers/utils'
 
 import type { ModelInfo, ProviderMetadata, VoiceInfo } from '../../providers'
 
+import { toWavFromPCM16 } from '@proj-airi/audio/encoding'
+
 import { OPENROUTER_ATTRIBUTION_HEADERS } from '../../../libs/providers/providers/openrouter-ai'
 
 const DEFAULT_BASE_URL = 'https://openrouter.ai/api/v1/'
@@ -94,42 +96,6 @@ function decodeBase64PCM(chunks: string[]): Uint8Array {
   return bytes
 }
 
-/** Wraps raw PCM16 mono data in a minimal WAV container. */
-function wrapPCM16InWAV(pcmBytes: Uint8Array, sampleRate = 24000): Uint8Array {
-  const numChannels = 1
-  const bitsPerSample = 16
-  const byteRate = sampleRate * numChannels * (bitsPerSample / 8)
-  const blockAlign = numChannels * (bitsPerSample / 8)
-  const header = new ArrayBuffer(44)
-  const view = new DataView(header)
-
-  const writeStr = (offset: number, str: string) => {
-    for (let i = 0; i < str.length; i++)
-      view.setUint8(offset + i, str.charCodeAt(i))
-  }
-
-  writeStr(0, 'RIFF')
-  view.setUint32(4, 36 + pcmBytes.length, true)
-  writeStr(8, 'WAVE')
-
-  writeStr(12, 'fmt ')
-  view.setUint32(16, 16, true)
-  view.setUint16(20, 1, true)
-  view.setUint16(22, numChannels, true)
-  view.setUint32(24, sampleRate, true)
-  view.setUint32(28, byteRate, true)
-  view.setUint16(32, blockAlign, true)
-  view.setUint16(34, bitsPerSample, true)
-
-  writeStr(36, 'data')
-  view.setUint32(40, pcmBytes.length, true)
-
-  const wav = new Uint8Array(44 + pcmBytes.length)
-  wav.set(new Uint8Array(header), 0)
-  wav.set(pcmBytes, 44)
-  return wav
-}
-
 /**
  * Custom fetch adapter that translates an OpenAI-compatible TTS request
  * into an OpenRouter streaming chat-completion with audio modality,
@@ -167,9 +133,9 @@ function createAudioFetch(apiKey: string, baseUrl: string, model: string) {
 
     const audioChunks = await collectAudioChunksFromSSE(sseResponse.body!)
     const pcmBytes = decodeBase64PCM(audioChunks)
-    const wavBytes = wrapPCM16InWAV(pcmBytes)
+    const wavBuffer = toWavFromPCM16(pcmBytes, 24000)
 
-    return new Response(new Blob([wavBytes.buffer as ArrayBuffer], { type: 'audio/wav' }), {
+    return new Response(new Blob([wavBuffer], { type: 'audio/wav' }), {
       status: 200,
       headers: { 'Content-Type': 'audio/wav' },
     })