add support for newer openai_whisper engine responses in whisper asr

fixes djmango#52

src/transcribe.ts

@@ -6,7 +6,7 @@ import { PayloadData, payloadGenerator, preprocessWhisperASRResponse } from "src
 import { StatusBar } from "./status";
 import { SupabaseClient } from "@supabase/supabase-js";
 import * as tus from "tus-js-client";
-import { WhisperASRSegment } from "./types/whisper-asr";
+import { WhisperASRResponse, WhisperASRSegment } from "./types/whisper-asr";
 
 type TranscriptionBackend = (file: TFile) => Promise<string>;
 
@@ -157,14 +157,17 @@ export class TranscriptionEngine {
                 const response = await requestUrl(options);
                 if (this.settings.debug) console.log("Raw response:", response);
 
-                const preprocessed = preprocessWhisperASRResponse(response.json);
+                // ASR_ENGINE=faster_whisper returns segments as an array. Preprocess it to match the standard.
+                const preprocessed = Array.isArray(response.json.segments[0])
+                    ? preprocessWhisperASRResponse(response.json) : response.json as WhisperASRResponse;
+
                 if (this.settings.debug) console.log("Preprocessed response:", preprocessed);
 
                 // Create segments for each word timestamp if word timestamps are available
                 const wordSegments = preprocessed.segments
                     .reduce((acc: components["schemas"]["TimestampedTextSegment"][], segment: WhisperASRSegment) => {
-                        if (segment.wordTimestamps) {
-                            acc.push(...segment.wordTimestamps.map(wordTimestamp => ({
+                        if (segment.words) {
+                            acc.push(...segment.words.map(wordTimestamp => ({
                                 start: wordTimestamp.start,
                                 end: wordTimestamp.end,
                                 text: wordTimestamp.word

src/types/whisper-asr.ts

@@ -87,7 +87,12 @@ export type WhisperASRSegment = {
     /**
      * An array of word-level timestamps, providing detailed timing for each word spoken in the segment.
      */
-    wordTimestamps: WhisperASRWordTimestamp[] | null;
+    words: WhisperASRWordTimestamp[] | null;
+
+    /**
+     * The unique identifier for the segment.
+     */
+    id: number;
 };
 
 /**
@@ -116,7 +121,7 @@ export type WhisperASRWordTimestamp = {
      * The model's confidence in the accuracy of this word transcription.
      * @example 0.941351592540741
      */
-    confidence: number;
+    probability: number;
 };
 
 

src/utils.ts

@@ -1,6 +1,6 @@
 /* Utility functions for Obsidian Transcript */
 import { App, FileSystemAdapter, getBlobArrayBuffer } from "obsidian";
-import { WhisperASRResponse, WhisperASRSegment } from "./types/whisper-asr";
+import { WhisperASRResponse, WhisperASRSegment, WhisperASRWordTimestamp } from "./types/whisper-asr";
 
 export const randomString = (length: number) => Array(length + 1).join((Math.random().toString(36) + '00000000000000000').slice(2, 18)).slice(0, length)
 export const getAllLinesFromFile = (cache: string) => cache.split(/\r?\n/)
@@ -74,15 +74,15 @@ export function preprocessWhisperASRResponse(rawResponse: any): WhisperASRRespon
                 avg_logprob: segment[7],
                 compression_ratio: segment[8],
                 no_speech_prob: segment[9],
-                wordTimestamps: null,
+                words: null,
             } as WhisperASRSegment;
             if (segment[10] !== null) { // easier to read than a ternary-destructured assignment
-                baseSegment.wordTimestamps = segment[10].map((wordTimestamp: any) => ({
+                baseSegment.words = segment[10].map((wordTimestamp: unknown[]) => ({
                     start: wordTimestamp[0],
                     end: wordTimestamp[1],
                     word: wordTimestamp[2],
-                    confidence: wordTimestamp[3],
-                }));
+                    probability: wordTimestamp[3],
+                } as WhisperASRWordTimestamp));
             }
             return baseSegment;
         })