@storyteller-platform/ghost-story 0.1.10 → 0.1.11

package/dist/index.cjs

@@ -72,6 +72,9 @@ __export(index_exports, {
   serviceCapabilities: () => import_audio.serviceCapabilities,
   setConversionMode: () => import_config2.setConversionMode,
   setTimingEnabled: () => import_config2.setTimingEnabled,
+  spacelessScriptPattern: () => import_SpacelessScripts.spacelessScriptPattern,
+  spacelessScripts: () => import_SpacelessScripts.spacelessScripts,
+  startsWithSpacelessScript: () => import_SpacelessScripts.startsWithSpacelessScript,
   toBuffer: () => import_audio.toBuffer,
   toFilePath: () => import_audio.toFilePath,
   toReadStream: () => import_audio.toReadStream,
@@ -87,6 +90,7 @@ var import_config2 = require("./config.cjs");
 var import_OpenAICloudSTT = require("./recognition/OpenAICloudSTT.cjs");
 var import_WhisperCppSTT = require("./recognition/WhisperCppSTT.cjs");
 var import_WhisperServerSTT = require("./recognition/WhisperServerSTT.cjs");
+var import_SpacelessScripts = require("./utilities/SpacelessScripts.cjs");
 var import_Timing = require("./utilities/Timing.cjs");
 var import_Silero = require("./vad/Silero.cjs");
 // Annotate the CommonJS export names for ESM import in node:
@@ -145,6 +149,9 @@ var import_Silero = require("./vad/Silero.cjs");
   serviceCapabilities,
   setConversionMode,
   setTimingEnabled,
+  spacelessScriptPattern,
+  spacelessScripts,
+  startsWithSpacelessScript,
   toBuffer,
   toFilePath,
   toReadStream,

package/dist/index.d.cts

@@ -9,6 +9,7 @@ export { OpenAICloudSTTOptions, RecognitionResult as OpenAIResult, inputPreferen
 export { WhisperCppModelId, WhisperCppOptions, RecognitionResult as WhisperCppResult, recognize as recognizeWhisperCpp, inputPreference as whisperCppInputPreference } from './recognition/WhisperCppSTT.cjs';
 export { WhisperServerOptions, RecognitionResult as WhisperServerResult, recognize as recognizeWhisperServer, inputPreference as whisperServerInputPreference } from './recognition/WhisperServerSTT.cjs';
 export { Timeline, TimelineEntry, TimelineEntryType } from './utilities/Timeline.cjs';
+export { spacelessScriptPattern, spacelessScripts, startsWithSpacelessScript } from './utilities/SpacelessScripts.cjs';
 export { AggregatedStats, PhaseTiming, Timing, TimingAggregator, TimingSpan, TimingSummary, createAggregator, createTiming, formatDuration, formatPercentage, formatSingleReport, printSingleReport } from './utilities/Timing.cjs';
 export { SileroOptions, VadSegment, detectVoiceActivity, ensureVadInstalled, segmentsToTimeline } from './vad/Silero.cjs';
 import 'node:fs';

package/dist/index.d.ts

@@ -9,6 +9,7 @@ export { OpenAICloudSTTOptions, RecognitionResult as OpenAIResult, inputPreferen
 export { WhisperCppModelId, WhisperCppOptions, RecognitionResult as WhisperCppResult, recognize as recognizeWhisperCpp, inputPreference as whisperCppInputPreference } from './recognition/WhisperCppSTT.js';
 export { WhisperServerOptions, RecognitionResult as WhisperServerResult, recognize as recognizeWhisperServer, inputPreference as whisperServerInputPreference } from './recognition/WhisperServerSTT.js';
 export { Timeline, TimelineEntry, TimelineEntryType } from './utilities/Timeline.js';
+export { spacelessScriptPattern, spacelessScripts, startsWithSpacelessScript } from './utilities/SpacelessScripts.js';
 export { AggregatedStats, PhaseTiming, Timing, TimingAggregator, TimingSpan, TimingSummary, createAggregator, createTiming, formatDuration, formatPercentage, formatSingleReport, printSingleReport } from './utilities/Timing.js';
 export { SileroOptions, VadSegment, detectVoiceActivity, ensureVadInstalled, segmentsToTimeline } from './vad/Silero.js';
 import 'node:fs';

package/dist/index.js

@@ -62,6 +62,11 @@ import {
   inputPreference as inputPreference3,
   recognize as recognize4
 } from "./recognition/WhisperServerSTT.js";
+import {
+  spacelessScriptPattern,
+  spacelessScripts,
+  startsWithSpacelessScript
+} from "./utilities/SpacelessScripts.js";
 import {
   Timing,
   TimingAggregator,
@@ -132,6 +137,9 @@ export {
   serviceCapabilities,
   setConversionMode,
   setTimingEnabled,
+  spacelessScriptPattern,
+  spacelessScripts,
+  startsWithSpacelessScript,
   toBuffer,
   toFilePath,
   toReadStream,

package/dist/recognition/WhisperServerSTT.cjs

@@ -59,11 +59,49 @@ async function recognize(input, languageCode, timing, options) {
     if (languageCode) {
       form.append("language", languageCode);
     }
-    const url = `${opts.baseURL}${opts.inferencePath}`;
+    const baseUrl = opts.baseURL.replace(/\/+$/g, "");
+    const url = `${baseUrl}${opts.inferencePath}`;
     const headers = {};
     if (opts.apiKey) {
       headers["Authorization"] = `Bearer ${opts.apiKey}`;
     }
+    const configResponse = await fetch(`${baseUrl}/config`, {
+      headers,
+      dispatcher: (0, import_fetch.createTimeoutAgent)(opts.timeout)
+    });
+    let whisperConfig = null;
+    if (configResponse.ok) {
+      try {
+        const [config, audioLength] = await Promise.all([
+          configResponse.json(),
+          (0, import_audio.getAudioDuration)(filePath)
+        ]);
+        whisperConfig = {
+          ...config,
+          audioDuration: audioLength
+        };
+        const effectiveProcessors = (0, import_WhisperTimeline.calculateEffectiveProcessors)(
+          audioLength,
+          whisperConfig.maxProcessors
+        );
+        if (effectiveProcessors !== whisperConfig.processors) {
+          const configForm = new FormData();
+          configForm.append("processors", String(effectiveProcessors));
+          configForm.append("threads", String(whisperConfig.threads));
+          await fetch(`${baseUrl}/config`, {
+            method: "POST",
+            headers,
+            body: configForm
+          });
+          whisperConfig.processors = effectiveProcessors;
+        }
+      } catch (e) {
+        console.warn(
+          `Failed to get config from Whisper server, continuing with default config. If you aren't using ghost-story server, this is expected`,
+          e
+        );
+      }
+    }
     const response = await timing.timeAsync(
       "upload",
       async () => fetch(url, {
@@ -81,7 +119,8 @@ async function recognize(input, languageCode, timing, options) {
     const data = await response.json();
     const { timeline, transcript } = await extractTimelineAndTranscript(
       data,
-      filePath
+      filePath,
+      whisperConfig
     );
     if (!timeline) {
       throw new Error(
@@ -93,7 +132,7 @@ async function recognize(input, languageCode, timing, options) {
     await prepared.cleanup();
   }
 }
-async function extractTimelineAndTranscript(response, audioPath) {
+async function extractTimelineAndTranscript(response, audioPath, whisperConfig) {
   var _a, _b, _c, _d;
   if (response.segments.length === 0) {
     return { timeline: [], transcript: ((_a = response.text) == null ? void 0 : _a.trim()) ?? "" };
@@ -101,7 +140,10 @@ async function extractTimelineAndTranscript(response, audioPath) {
   const hasNestedWords = (((_c = (_b = response.segments[0]) == null ? void 0 : _b.words) == null ? void 0 : _c.length) ?? 0) > 0;
   if (hasNestedWords) {
     const rawSegments = (0, import_WhisperTimeline.parseWhisperServerOutput)(response.segments);
-    const splitBoundaries = await detectSplitBoundaries(rawSegments, audioPath);
+    const splitBoundaries = (whisperConfig == null ? void 0 : whisperConfig.audioDuration) ? (0, import_WhisperTimeline.calculateWhisperSplits)(
+      whisperConfig.audioDuration,
+      whisperConfig.processors
+    ) : await detectSplitBoundaries(rawSegments, audioPath);
     const timeline2 = (0, import_WhisperTimeline.extractCorrectedTimeline)(rawSegments, { splitBoundaries });
     const transcript = timeline2.map((entry) => entry.text).join(" ");
     return { timeline: timeline2, transcript };

package/dist/recognition/WhisperServerSTT.js

@@ -9,6 +9,7 @@ import {
 } from "../audio/index.js";
 import { createTimeoutAgent } from "../fetch.js";
 import {
+  calculateEffectiveProcessors,
   calculateWhisperSplits,
   countProcessorBoundaries,
   extractCorrectedTimeline,
@@ -46,11 +47,49 @@ async function recognize(input, languageCode, timing, options) {
     if (languageCode) {
       form.append("language", languageCode);
     }
-    const url = `${opts.baseURL}${opts.inferencePath}`;
+    const baseUrl = opts.baseURL.replace(/\/+$/g, "");
+    const url = `${baseUrl}${opts.inferencePath}`;
     const headers = {};
     if (opts.apiKey) {
       headers["Authorization"] = `Bearer ${opts.apiKey}`;
     }
+    const configResponse = await fetch(`${baseUrl}/config`, {
+      headers,
+      dispatcher: createTimeoutAgent(opts.timeout)
+    });
+    let whisperConfig = null;
+    if (configResponse.ok) {
+      try {
+        const [config, audioLength] = await Promise.all([
+          configResponse.json(),
+          getAudioDuration(filePath)
+        ]);
+        whisperConfig = {
+          ...config,
+          audioDuration: audioLength
+        };
+        const effectiveProcessors = calculateEffectiveProcessors(
+          audioLength,
+          whisperConfig.maxProcessors
+        );
+        if (effectiveProcessors !== whisperConfig.processors) {
+          const configForm = new FormData();
+          configForm.append("processors", String(effectiveProcessors));
+          configForm.append("threads", String(whisperConfig.threads));
+          await fetch(`${baseUrl}/config`, {
+            method: "POST",
+            headers,
+            body: configForm
+          });
+          whisperConfig.processors = effectiveProcessors;
+        }
+      } catch (e) {
+        console.warn(
+          `Failed to get config from Whisper server, continuing with default config. If you aren't using ghost-story server, this is expected`,
+          e
+        );
+      }
+    }
     const response = await timing.timeAsync(
       "upload",
       async () => fetch(url, {
@@ -68,7 +107,8 @@ async function recognize(input, languageCode, timing, options) {
     const data = await response.json();
     const { timeline, transcript } = await extractTimelineAndTranscript(
       data,
-      filePath
+      filePath,
+      whisperConfig
     );
     if (!timeline) {
       throw new Error(
@@ -80,7 +120,7 @@ async function recognize(input, languageCode, timing, options) {
     await prepared.cleanup();
   }
 }
-async function extractTimelineAndTranscript(response, audioPath) {
+async function extractTimelineAndTranscript(response, audioPath, whisperConfig) {
   var _a, _b, _c, _d;
   if (response.segments.length === 0) {
     return { timeline: [], transcript: ((_a = response.text) == null ? void 0 : _a.trim()) ?? "" };
@@ -88,7 +128,10 @@ async function extractTimelineAndTranscript(response, audioPath) {
   const hasNestedWords = (((_c = (_b = response.segments[0]) == null ? void 0 : _b.words) == null ? void 0 : _c.length) ?? 0) > 0;
   if (hasNestedWords) {
     const rawSegments = parseWhisperServerOutput(response.segments);
-    const splitBoundaries = await detectSplitBoundaries(rawSegments, audioPath);
+    const splitBoundaries = (whisperConfig == null ? void 0 : whisperConfig.audioDuration) ? calculateWhisperSplits(
+      whisperConfig.audioDuration,
+      whisperConfig.processors
+    ) : await detectSplitBoundaries(rawSegments, audioPath);
     const timeline2 = extractCorrectedTimeline(rawSegments, { splitBoundaries });
     const transcript = timeline2.map((entry) => entry.text).join(" ");
     return { timeline: timeline2, transcript };

package/dist/utilities/WhisperTimeline.cjs

@@ -302,14 +302,28 @@ function extractCorrectedTimeline(segments, options = {}) {
     if (!segment) continue;
     const segmentStart = segment.segmentStart;
     const segmentEnd = segment.segmentEnd < segment.segmentStart ? segment.segmentStart : segment.segmentEnd;
-    const boundary = detectProcessorBoundary(segment, state);
-    if (boundary.isBoundary) {
-      state.cumulativeOffset = getBetterCumulativeOffset(
-        state,
-        segment,
-        splitBoundaries,
-        usedSplits
-      );
+    if (splitBoundaries.length > 0) {
+      const boundary = splitBoundaries.find((boundary2) => {
+        return Math.abs(boundary2 - segmentStart) < 2;
+      });
+      if (boundary) {
+        state.cumulativeOffset = getBetterCumulativeOffset(
+          state,
+          segment,
+          splitBoundaries,
+          usedSplits
+        );
+      }
+    } else {
+      const boundary = detectProcessorBoundary(segment, state);
+      if (boundary.isBoundary) {
+        state.cumulativeOffset = getBetterCumulativeOffset(
+          state,
+          segment,
+          splitBoundaries,
+          usedSplits
+        );
+      }
     }
     const nextSegment = segments[i + 1] ?? null;
     if (nextSegment && isTimeTravelingSegment(nextSegment)) {

package/dist/utilities/WhisperTimeline.js

@@ -272,14 +272,28 @@ function extractCorrectedTimeline(segments, options = {}) {
     if (!segment) continue;
     const segmentStart = segment.segmentStart;
     const segmentEnd = segment.segmentEnd < segment.segmentStart ? segment.segmentStart : segment.segmentEnd;
-    const boundary = detectProcessorBoundary(segment, state);
-    if (boundary.isBoundary) {
-      state.cumulativeOffset = getBetterCumulativeOffset(
-        state,
-        segment,
-        splitBoundaries,
-        usedSplits
-      );
+    if (splitBoundaries.length > 0) {
+      const boundary = splitBoundaries.find((boundary2) => {
+        return Math.abs(boundary2 - segmentStart) < 2;
+      });
+      if (boundary) {
+        state.cumulativeOffset = getBetterCumulativeOffset(
+          state,
+          segment,
+          splitBoundaries,
+          usedSplits
+        );
+      }
+    } else {
+      const boundary = detectProcessorBoundary(segment, state);
+      if (boundary.isBoundary) {
+        state.cumulativeOffset = getBetterCumulativeOffset(
+          state,
+          segment,
+          splitBoundaries,
+          usedSplits
+        );
+      }
     }
     const nextSegment = segments[i + 1] ?? null;
     if (nextSegment && isTimeTravelingSegment(nextSegment)) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@storyteller-platform/ghost-story",
-  "version": "0.1.10",
+  "version": "0.1.11",
   "description": "An easy-to-use speech toolset. Fork of the original echogarden project.",
   "author": "Thomas F. K. Jorna",
   "license": "GPL-3.0",