feat(generate-transcription): opt-in verbose mode (#159)

* feat(generate-transcription): opt-in verbose mode

* chore(generate-transcription): update test

* chore(generate-transcription): clean snapshot

packages/generate-transcription/src/index.ts

@@ -2,24 +2,42 @@ import type { CommonRequestOptions } from '@xsai/shared'
 
 import { requestHeaders, requestURL, responseJSON } from '@xsai/shared'
 
-export interface GenerateTranscriptionOptions<T extends GenerateTranscriptionOptionsTimeStampGranularities = undefined> extends CommonRequestOptions {
+export interface GenerateTranscriptionOptions<
+  T1 extends GenerateTranscriptionOptionsResponseFormat,
+  T2 extends GenerateTranscriptionOptionsTimeStampGranularities,
+> extends CommonRequestOptions {
   file: Blob
   fileName?: string
   language?: string
   prompt?: string
+  /** @default `json` */
+  responseFormat?: T1
   temperature?: string
   /** @default `segment` */
-  timestampGranularities?: T
+  timestampGranularities?: T2
 }
 
+export type GenerateTranscriptionOptionsResponseFormat = 'json' | 'verbose_json' | undefined
+
 export type GenerateTranscriptionOptionsTimeStampGranularities = 'segment' | 'word' | undefined
 
-export interface GenerateTranscriptionResult<T extends GenerateTranscriptionOptionsTimeStampGranularities = undefined> {
-  duration: number
-  language: string
-  segments: T extends 'word' ? never : GenerateTranscriptionResultSegment[]
+export interface GenerateTranscriptionResult<
+  T1 extends GenerateTranscriptionOptionsResponseFormat,
+  T2 extends GenerateTranscriptionOptionsTimeStampGranularities,
+> {
+  duration: T1 extends 'verbose_json' ? number : never
+  language: T1 extends 'verbose_json' ? string : never
+  segments: T1 extends 'verbose_json'
+    ? T2 extends 'word'
+      ? never
+      : GenerateTranscriptionResultSegment[]
+    : never
   text: string
-  words: T extends 'word' ? GenerateTranscriptionResultWord[] : never
+  words: T1 extends 'verbose_json'
+    ? T2 extends 'word'
+      ? GenerateTranscriptionResultWord[]
+      : never
+    : never
 }
 
 /** @see {@link https://platform.openai.com/docs/api-reference/audio/verbose-json-object#audio/verbose-json-object-segments} */
@@ -43,13 +61,20 @@ export interface GenerateTranscriptionResultWord {
   word: string
 }
 
-export const generateTranscription = async <T extends GenerateTranscriptionOptionsTimeStampGranularities = undefined>(options: GenerateTranscriptionOptions<T>): Promise<GenerateTranscriptionResult<T>> => {
+export const generateTranscription = async <
+  T1 extends GenerateTranscriptionOptionsResponseFormat = undefined,
+  T2 extends GenerateTranscriptionOptionsTimeStampGranularities = undefined,
+>(options: GenerateTranscriptionOptions<T1, T2>): Promise<GenerateTranscriptionResult<T1, T2>> => {
   const body = new FormData()
 
   body.append('model', options.model)
   body.append('file', options.file, options.fileName)
-  body.append('response_format', 'verbose_json')
-  body.append('timestamp_granularities[]', options.timestampGranularities ?? 'segment')
+  body.append('response_format', options.responseFormat ?? 'json')
+
+  // make ts happy
+  // eslint-disable-next-line ts/no-unnecessary-type-assertion
+  if (options.responseFormat as GenerateTranscriptionOptionsResponseFormat === 'verbose_json')
+    body.append('timestamp_granularities[]', options.timestampGranularities ?? 'segment')
 
   if (options.language != null)
     body.append('language', options.language)
@@ -66,5 +91,5 @@ export const generateTranscription = async <T extends GenerateTranscriptionOptio
     method: 'POST',
     signal: options.abortSignal,
   })
-    .then(responseJSON<GenerateTranscriptionResult<T>>)
+    .then(responseJSON<GenerateTranscriptionResult<T1, T2>>)
 }

packages/generate-transcription/test/__snapshots__/index.test.ts.snap

@@ -1,6 +1,6 @@
 // Vitest Snapshot v1, https://vitest.dev/guide/snapshot.html
 
-exports[`@xsai/generate-transcription > segment 1`] = `
+exports[`@xsai/generate-transcription > verbose_json + segment 1`] = `
 [
   {
     "avg_logprob": -0.084375,
@@ -43,7 +43,7 @@ exports[`@xsai/generate-transcription > segment 1`] = `
 ]
 `;
 
-exports[`@xsai/generate-transcription > word 1`] = `
+exports[`@xsai/generate-transcription > verbose_json + word 1`] = `
 [
   {
     "end": 0.4,

packages/generate-transcription/test/index.test.ts

@@ -4,7 +4,22 @@ import { describe, expect, it } from 'vitest'
 import { generateTranscription } from '../src'
 
 describe('@xsai/generate-transcription', () => {
-  it('segment', async () => {
+  const expectText = 'Hello, I am your AI assistant. Just let me know how I can help bring your ideas to life.'
+
+  it('json', async () => {
+    const { text } = await generateTranscription({
+      apiKey: 'a',
+      baseURL: new URL('http://localhost:8000/v1/'),
+      file: await openAsBlob('./test/fixtures/basic.wav', { type: 'audio/wav' }),
+      fileName: 'basic.wav',
+      language: 'en',
+      model: 'deepdml/faster-whisper-large-v3-turbo-ct2',
+    })
+
+    expect(text).toBe(expectText)
+  })
+
+  it('verbose_json + segment', async () => {
     const { duration, language, segments, text } = await generateTranscription({
       apiKey: 'a',
       baseURL: new URL('http://localhost:8000/v1/'),
@@ -12,15 +27,16 @@ describe('@xsai/generate-transcription', () => {
       fileName: 'basic.wav',
       language: 'en',
       model: 'deepdml/faster-whisper-large-v3-turbo-ct2',
+      responseFormat: 'verbose_json',
     })
 
     expect(duration).toBe(5.472)
     expect(language).toBe('en')
-    expect(text).toBe('Hello, I am your AI assistant. Just let me know how I can help bring your ideas to life.')
+    expect(text).toBe(expectText)
     expect(segments).toMatchSnapshot()
   }, 30000)
 
-  it('word', async () => {
+  it('verbose_json + word', async () => {
     const { duration, language, text, words } = await generateTranscription({
       apiKey: 'a',
       baseURL: new URL('http://localhost:8000/v1/'),
@@ -28,12 +44,13 @@ describe('@xsai/generate-transcription', () => {
       fileName: 'basic.wav',
       language: 'en',
       model: 'deepdml/faster-whisper-large-v3-turbo-ct2',
+      responseFormat: 'verbose_json',
       timestampGranularities: 'word',
     })
 
     expect(duration).toBe(5.472)
     expect(language).toBe('en')
-    expect(text).toBe('Hello, I am your AI assistant. Just let me know how I can help bring your ideas to life.')
+    expect(text).toBe(expectText)
     expect(words).toMatchSnapshot()
   }, 30000)
 })