130 lines
4.8 KiB
TypeScript
130 lines
4.8 KiB
TypeScript
import { FileSystem } from 'react-native-file-access'
|
|
import type { RecordedAudio } from '../src/lib/audio-recorder'
|
|
import { AudioPipelineError, type LocalAudioInput } from '../src/features/import/audio-import-types'
|
|
import { prepareRecordedAudio } from '../src/features/import/recorded-audio-input'
|
|
import { transcribeAudioLocally } from '../src/features/import/local-whisper-transcription'
|
|
import { transcribeTalkRecording } from '../src/features/talk/talk-transcription-service'
|
|
|
|
jest.mock('../src/features/import/recorded-audio-input', () => ({
|
|
prepareRecordedAudio: jest.fn(),
|
|
}))
|
|
jest.mock('../src/features/import/local-whisper-transcription', () => ({
|
|
transcribeAudioLocally: jest.fn(),
|
|
}))
|
|
|
|
const originalFetch = global.fetch
|
|
const recording: RecordedAudio = {
|
|
uri: 'file:///cache/talk.m4a',
|
|
path: '/cache/talk.m4a',
|
|
fileName: 'talk.m4a',
|
|
mimeType: 'audio/mp4',
|
|
size: 3,
|
|
durationMs: 1_500,
|
|
}
|
|
|
|
function localInput(dispose: jest.Mock): LocalAudioInput {
|
|
return {
|
|
uri: recording.uri,
|
|
path: recording.path,
|
|
fileName: recording.fileName,
|
|
mimeType: recording.mimeType,
|
|
sizeBytes: recording.size,
|
|
durationMs: recording.durationMs,
|
|
source: 'recording',
|
|
dispose,
|
|
}
|
|
}
|
|
|
|
beforeEach(() => {
|
|
jest.clearAllMocks()
|
|
;(FileSystem.readFile as jest.Mock).mockResolvedValue('AQID')
|
|
})
|
|
|
|
afterEach(() => {
|
|
global.fetch = originalFetch
|
|
})
|
|
|
|
describe('Talk speech transcription', () => {
|
|
it('uses a real authenticated stt-proxy response and always removes the recording', async () => {
|
|
const dispose = jest.fn().mockResolvedValue(undefined)
|
|
;(prepareRecordedAudio as jest.Mock).mockResolvedValue(localInput(dispose))
|
|
global.fetch = jest.fn().mockResolvedValue({
|
|
ok: true,
|
|
status: 200,
|
|
json: async () => ({
|
|
transcript: '실제 음성',
|
|
confidence: 0.9,
|
|
language_code: 'ko',
|
|
duration_seconds: 1.5,
|
|
provider: 'deepgram',
|
|
}),
|
|
})
|
|
|
|
await expect(transcribeTalkRecording(recording, {
|
|
accessToken: 'user-token',
|
|
languageCode: 'ko',
|
|
signal: new AbortController().signal,
|
|
disposeRecording: jest.fn(),
|
|
})).resolves.toMatchObject({ text: '실제 음성', provider: 'deepgram' })
|
|
expect(dispose).toHaveBeenCalledTimes(1)
|
|
expect(transcribeAudioLocally).not.toHaveBeenCalled()
|
|
expect((global.fetch as jest.Mock).mock.calls[0][1].headers.Authorization)
|
|
.toBe('Bearer user-token')
|
|
})
|
|
|
|
it('falls back to the bundled on-device Whisper only for cloud/provider failures', async () => {
|
|
const dispose = jest.fn().mockResolvedValue(undefined)
|
|
const input = localInput(dispose)
|
|
;(prepareRecordedAudio as jest.Mock).mockResolvedValue(input)
|
|
;(transcribeAudioLocally as jest.Mock).mockResolvedValue({
|
|
text: '로컬 전사',
|
|
confidence: null,
|
|
language: 'ko',
|
|
durationSeconds: 1.5,
|
|
provider: 'whisper.cpp-tiny-local',
|
|
latencyMs: 120,
|
|
})
|
|
global.fetch = jest.fn().mockResolvedValue({ ok: false, status: 503 })
|
|
|
|
await expect(transcribeTalkRecording(recording, {
|
|
accessToken: 'user-token',
|
|
languageCode: 'ko',
|
|
signal: new AbortController().signal,
|
|
disposeRecording: jest.fn(),
|
|
})).resolves.toMatchObject({ text: '로컬 전사', provider: 'whisper.cpp-tiny-local' })
|
|
expect(transcribeAudioLocally).toHaveBeenCalledWith(input, 'ko', expect.any(AbortSignal))
|
|
expect(dispose).toHaveBeenCalledTimes(1)
|
|
})
|
|
|
|
it('does not bypass authentication or quota failures with local transcription', async () => {
|
|
const dispose = jest.fn().mockResolvedValue(undefined)
|
|
;(prepareRecordedAudio as jest.Mock).mockResolvedValue(localInput(dispose))
|
|
global.fetch = jest.fn().mockResolvedValue({ ok: false, status: 401 })
|
|
|
|
await expect(transcribeTalkRecording(recording, {
|
|
accessToken: 'expired-token',
|
|
languageCode: 'ko',
|
|
signal: new AbortController().signal,
|
|
disposeRecording: jest.fn(),
|
|
})).rejects.toMatchObject({ code: 'auth' })
|
|
expect(transcribeAudioLocally).not.toHaveBeenCalled()
|
|
expect(dispose).toHaveBeenCalledTimes(1)
|
|
})
|
|
|
|
it('reports both cloud and local failures without inventing a transcript', async () => {
|
|
const dispose = jest.fn().mockResolvedValue(undefined)
|
|
;(prepareRecordedAudio as jest.Mock).mockResolvedValue(localInput(dispose))
|
|
;(transcribeAudioLocally as jest.Mock).mockRejectedValue(
|
|
new AudioPipelineError('transcription', 'Whisper decoded no speech'),
|
|
)
|
|
global.fetch = jest.fn().mockRejectedValue(new TypeError('network down'))
|
|
|
|
await expect(transcribeTalkRecording(recording, {
|
|
accessToken: 'user-token',
|
|
languageCode: 'ko',
|
|
signal: new AbortController().signal,
|
|
disposeRecording: jest.fn(),
|
|
})).rejects.toMatchObject({ code: 'transcription', message: 'Whisper decoded no speech' })
|
|
expect(dispose).toHaveBeenCalledTimes(1)
|
|
})
|
|
})
|