omnius 1.0.427 → 1.0.429

package/dist/index.js

@@ -603384,6 +603384,15 @@ async function findLiveWhisperScript() {
   }
   return null;
 }
+function defaultConsensusModel(primaryModel) {
+  const model = primaryModel.trim().toLowerCase();
+  if (model === "tiny") return "base";
+  if (model === "base") return "small";
+  if (model === "small") return "base";
+  if (model === "medium") return "base";
+  if (model === "large-v3" || model === "large") return "small";
+  return "base";
+}
 async function ensureVenvForTranscribeCli() {
   const bin = process.platform === "win32" ? "Scripts" : "bin";
   const exe = process.platform === "win32" ? "python.exe" : "python3";
@@ -603573,7 +603582,9 @@ var init_listen = __esm({
           lastStatus: `loading whisper ${this.model} model...`
         });
         const consensusEnv = String(process.env["OMNIUS_ASR_CONSENSUS"] ?? "").trim().toLowerCase();
-        const consensusModel = consensusEnv === "0" || consensusEnv === "off" || consensusEnv === "false" ? "off" : consensusEnv || (this.model === "tiny" ? "base" : "tiny");
+        const consensusModel = consensusEnv === "0" || consensusEnv === "off" || consensusEnv === "false" ? "off" : consensusEnv || defaultConsensusModel(this.model);
+        const consensusThreshold = String(process.env["OMNIUS_ASR_CONSENSUS_THRESHOLD"] ?? "0.45");
+        const silenceMs = String(process.env["OMNIUS_ASR_SILENCE_MS"] ?? "800");
         this.process = spawn29(
           pyPath,
           [
@@ -603585,7 +603596,11 @@ var init_listen = __esm({
             "--window-seconds",
             "10",
             "--consensus-model",
-            consensusModel
+            consensusModel,
+            "--consensus-threshold",
+            consensusThreshold,
+            "--silence-ms",
+            silenceMs
           ],
           {
             stdio: ["pipe", "pipe", "pipe"],
@@ -603625,13 +603640,18 @@ var init_listen = __esm({
                   });
                   break;
                 case "consensus_rejected":
-                  updateListenLiveState({
-                    lastStatus: `consensus rejected (${evt.reason ?? "mismatch"}): ${String(evt.primary ?? "").slice(0, 80)}`
-                  });
+                  {
+                    const message2 = `consensus rejected (${evt.reason ?? "mismatch"}): ${String(evt.primary ?? "").slice(0, 80)}`;
+                    updateListenLiveState({
+                      lastStatus: message2
+                    });
+                    this.emit("status", message2);
+                  }
                   break;
                 case "vad": {
                   lastHardwareVadAt = Date.now();
                   const doa = Number(evt.doa);
+                  this.emit("status", `Whisper VAD: ${evt.speech === true ? "speech" : "silence"}`);
                   updateListenLiveState({
                     speechActive: evt.speech === true,
                     doa: Number.isFinite(doa) ? doa : null
@@ -604033,8 +604053,9 @@ transcribe-cli error: ${transcribeCliError}` : "";
             return `Failed to start live transcription: ${msg}${tcHint}`;
           }
         }
-        this.spawnMicProcess(micCmd);
         this.active = true;
+        this.paused = false;
+        this.spawnMicProcess(micCmd);
         this.blinkState = true;
         this.blinkTimer = setInterval(() => {
           this.blinkState = !this.blinkState;
@@ -606454,6 +606475,7 @@ var init_live_sensors = __esm({
       previewTickInFlight = false;
       lastStreamFrameMtime = /* @__PURE__ */ new Map();
       lastStreamPersistAt = 0;
+      videoGeneration = 0;
       videoInFlight = false;
       audioInFlight = false;
       audioActivityInFlight = false;
@@ -606656,7 +606678,12 @@ var init_live_sensors = __esm({
         const next = (current + delta) % 360;
         return this.setCameraRotation(device, next, "manual", `manual ${direction === "cw" ? "clockwise" : "counter-clockwise"} dashboard override`);
       }
+      isVideoWorkCurrent(generation) {
+        return this.config.videoEnabled && generation === this.videoGeneration;
+      }
       async autoOrientCamera(device = this.config.selectedCamera, options2 = {}) {
+        const generation = this.videoGeneration;
+        if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
         const target = device || this.config.selectedCamera || firstDeviceId(this.devices.video);
         if (!target) return "No camera selected. Use /live camera <device> or select a camera in /live.";
         if (options2.onlyIfUnset && this.getCameraOrientation(target)) {
@@ -606673,17 +606700,20 @@ var init_live_sensors = __esm({
             framePath = streamFrame.path;
           }
           if (!framePath) {
+            if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
             const captured = await new CameraCaptureTool().execute({
               action: "capture",
               device: target,
               rotate_degrees: 0
             });
+            if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
             if (!captured.success) {
               return `Camera orientation detection could not capture ${target}: ${captured.error || captured.output || "unknown error"}`;
             }
             framePath = extractSavedImagePath(captured.output, this.repoRoot);
           }
           if (!framePath) return `Camera orientation detection captured ${target}, but no frame path was returned.`;
+          if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
           const cvScores = await scoreCameraOrientationWithOpenCv(framePath);
           const cvDecision = cvScores ? chooseCameraOrientationFromScores(cvScores) : null;
           const prompt = [
@@ -606701,6 +606731,7 @@ var init_live_sensors = __esm({
             prompt,
             model: "moondream3-preview"
           });
+          if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
           if (!vision.success) {
             if (cvDecision) {
               const orientation2 = this.setCameraRotation(target, cvDecision.rotation, "auto", cvDecision.reason, cvDecision.confidence);
@@ -606754,12 +606785,18 @@ var init_live_sensors = __esm({
         return this.devices;
       }
       configure(patch) {
+        const wasVideoEnabled = this.config.videoEnabled;
         this.config = {
           ...this.config,
           ...patch,
           videoIntervalMs: Math.max(MIN_VIDEO_INTERVAL_MS, Number(patch.videoIntervalMs ?? this.config.videoIntervalMs)),
           audioIntervalMs: Math.max(MIN_AUDIO_INTERVAL_MS, Number(patch.audioIntervalMs ?? this.config.audioIntervalMs))
         };
+        if (wasVideoEnabled && !this.config.videoEnabled) {
+          this.videoGeneration++;
+          this.lastStreamFrameMtime.clear();
+          this.cameraStreamers?.stopAll();
+        }
         this.persist();
         this.clearLoopTimers();
         this.ensureLoops();
@@ -606838,6 +606875,8 @@ var init_live_sensors = __esm({
         return options2.agent ? "Live PFC run enabled: visual/audio feedback is active and significant events can launch background agent review." : "Live run enabled: visual/audio feedback is active and live context will update each turn.";
       }
       async previewCamera(device = this.config.selectedCamera, options2 = {}) {
+        const generation = options2.generation ?? this.videoGeneration;
+        if (!this.isVideoWorkCurrent(generation)) return { ok: false, message: "Video context is disabled." };
         const source = device || firstDeviceId(this.devices.video);
         if (!source) return { ok: false, message: "No camera selected. Use /live to select a camera after device discovery." };
         const orientation = this.getCameraOrientation(source);
@@ -606847,6 +606886,7 @@ var init_live_sensors = __esm({
           const displayPath2 = relative11(this.repoRoot, streamFrame.path).startsWith("..") ? streamFrame.path : relative11(this.repoRoot, streamFrame.path);
           const previewWidth2 = Math.max(42, Math.min(120, Math.floor(Number(options2.previewWidth ?? (process.stdout.columns || 100) - 14))));
           const preview2 = await buildImageAsciiPreview(streamFrame.path, { width: previewWidth2 });
+          if (!this.isVideoWorkCurrent(generation)) return { ok: false, message: "Video context is disabled." };
           if (preview2?.ascii || preview2?.plainAscii) {
             const observedAt2 = Date.now();
             const cameraView2 = {
@@ -606889,6 +606929,7 @@ var init_live_sensors = __esm({
           device: source,
           rotate_degrees: rotation
         });
+        if (!this.isVideoWorkCurrent(generation)) return { ok: false, message: "Video context is disabled." };
         if (!result.success) {
           const error = result.error || result.output || "camera capture failed";
           const observedAt2 = Date.now();
@@ -606933,6 +606974,7 @@ var init_live_sensors = __esm({
         const displayPath = relative11(this.repoRoot, framePath).startsWith("..") ? framePath : relative11(this.repoRoot, framePath);
         const previewWidth = Math.max(42, Math.min(120, Math.floor(Number(options2.previewWidth ?? (process.stdout.columns || 100) - 14))));
         const preview = await buildImageAsciiPreview(framePath, { width: previewWidth });
+        if (!this.isVideoWorkCurrent(generation)) return { ok: false, message: "Video context is disabled." };
         const asciiContext = preview ? formatImageAsciiContext(preview, displayPath) : void 0;
         const observedAt = Date.now();
         const cameraView = {
@@ -607041,7 +607083,10 @@ var init_live_sensors = __esm({
         }
       }
       async sampleSingleVideoNow(source, forceInfer, previewWidth, options2 = {}) {
-        const preview = await this.previewCamera(source, { emit: false, previewWidth });
+        const generation = options2.generation ?? this.videoGeneration;
+        if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
+        const preview = await this.previewCamera(source, { emit: false, previewWidth, generation });
+        if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
         void this.runLiveVlm(source, preview.ok ? preview.framePath : void 0).catch(() => {
         });
         let inference = "";
@@ -607076,6 +607121,7 @@ var init_live_sensors = __esm({
             crop_faces: true,
             recognize_faces: true
           });
+          if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
           inference = result.success ? String(result.output || result.llmContent || "") : `Video inference failed: ${result.error || result.output || "unknown error"}`;
           const livePayload = result.success ? parseLiveMediaPayload(result.llmContent || result.output) : null;
           if (livePayload) livePayload.source = source;
@@ -607148,6 +607194,7 @@ var init_live_sensors = __esm({
         if (this.config.clipEnabled && preview.ok && preview.framePath && source && sampledAt - lastClip >= 5e3) {
           this.lastClipAt.set(source, sampledAt);
           clipSummary = await this.recognizeFrameObjects(preview.framePath, source, sampledAt, { emit: false });
+          if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
           const current = this.snapshot.cameras?.[source];
           if (current) {
             const clipLine = formatClipDashboardSummary(clipSummary);
@@ -607176,9 +607223,12 @@ ${clipSummary}` : ""].filter(Boolean).join("\n");
         return result;
       }
       async sampleVideoNow(forceInfer = this.config.inferEnabled, options2 = {}) {
+        const generation = options2.generation ?? this.videoGeneration;
+        if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
         if (this.videoInFlight) return "Video sampler is already running.";
         this.videoInFlight = true;
         try {
+          if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
           const allSources = this.activeVideoSources();
           if (allSources.length === 0) return "No camera selected. Use /live refresh after connecting hardware.";
           const sources = options2.mode === "round-robin" ? [allSources[this.nextVideoSourceIndex++ % allSources.length]] : allSources;
@@ -607186,8 +607236,10 @@ ${clipSummary}` : ""].filter(Boolean).join("\n");
           const outputs = [];
           const previewWidth = liveDashboardPreviewWidthForSources(allSources.length);
           for (const source of sources) {
-            outputs.push(await this.sampleSingleVideoNow(source, forceInfer, previewWidth, options2));
+            if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
+            outputs.push(await this.sampleSingleVideoNow(source, forceInfer, previewWidth, { ...options2, generation }));
           }
+          if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
           this.emitDashboard();
           return outputs.map((output, index) => `## Camera ${sources[index]}
 ${output}`).join("\n\n");
@@ -697778,7 +697830,7 @@ Rules:
         this.emit("snr", { score: this.lastSignalScore });
         this.onPartialTranscript(text2);
         if (this.silenceTimer) clearTimeout(this.silenceTimer);
-        const waitMs = this._vadSilenceMs ?? VAD_SILENCE_MS;
+        const waitMs = isFinal ? 0 : this._vadSilenceMs ?? VAD_SILENCE_MS;
         this.silenceTimer = setTimeout(() => {
           if (this._state === "CAPTURING") {
             this.finalizeSegment();
@@ -697945,35 +697997,35 @@ ${toolOutput}` });
             const reply = extractVoiceModelReply(stripToolJsonLines(response.trim()));
             const finalSpoken = reply.text;
             if (!finalSpoken || reply.action === "silent") {
-              return;
-            }
-            if (typeof this.voice.waitUntilIdle === "function") {
-              try {
-                await this.voice.waitUntilIdle();
-              } catch {
+            } else {
+              if (typeof this.voice.waitUntilIdle === "function") {
+                try {
+                  await this.voice.waitUntilIdle();
+                } catch {
+                }
               }
-            }
-            this.context.push({ role: "assistant", content: finalSpoken });
-            this.trimContext();
-            this.setState("SPEAKING");
-            this.onAgentSpeech(finalSpoken);
-            this.lastAgentSpeech = { text: finalSpoken, startedAt: Date.now() };
-            this.voice.speak(finalSpoken);
-            this.voiceTranscript.push({ role: "assistant", content: finalSpoken, ts: Date.now() });
-            if (this.runner && this.turnCount % SUMMARY_INJECTION_INTERVAL === 0) {
-              this.injectSummary();
-            }
-            if (typeof this.voice.waitUntilIdle === "function") {
-              try {
-                await this.voice.waitUntilIdle();
-              } catch {
+              this.context.push({ role: "assistant", content: finalSpoken });
+              this.trimContext();
+              this.setState("SPEAKING");
+              this.onAgentSpeech(finalSpoken);
+              this.lastAgentSpeech = { text: finalSpoken, startedAt: Date.now() };
+              this.voice.speak(finalSpoken);
+              this.voiceTranscript.push({ role: "assistant", content: finalSpoken, ts: Date.now() });
+              if (this.runner && this.turnCount % SUMMARY_INJECTION_INTERVAL === 0) {
+                this.injectSummary();
+              }
+              if (typeof this.voice.waitUntilIdle === "function") {
+                try {
+                  await this.voice.waitUntilIdle();
+                } catch {
+                }
+              } else {
+                const estimatedMs = Math.max(1500, finalSpoken.length / 5 * (6e4 / 150));
+                await new Promise((r2) => setTimeout(r2, estimatedMs));
+              }
+              if (reply.action === "hangup") {
+                this.active = false;
               }
-            } else {
-              const estimatedMs = Math.max(1500, finalSpoken.length / 5 * (6e4 / 150));
-              await new Promise((r2) => setTimeout(r2, estimatedMs));
-            }
-            if (reply.action === "hangup") {
-              this.active = false;
             }
           }
         } catch (err) {
@@ -737140,6 +737192,11 @@ Respond concisely and safely. Remember: you are talking to the general public.`;
       listenEng.on("level", (evt) => {
         statusBar.setMicActivity(evt.speechActive, evt.levelDb);
       });
+      listenEng.on("info", (msg) => {
+        if (/consensus rejected|loading whisper|whisper model loaded|Whisper VAD|no accepted speech text|very short utterance/i.test(msg)) {
+          writeContent(() => renderInfo(`[voicechat/asr] ${msg}`));
+        }
+      });
       const summaryRunner = {
         injectUserMessage(content) {
           if (activeTask?.runner) {

package/dist/scripts/live-whisper.py

@@ -443,13 +443,18 @@ def main():
             result = run_transcribe(model, amplify(samples))
             text = _segment_filtered_text(result)
             if text:
-                emit_transcript(text, is_final=False, consensus=False)
+                emit_transcript(
+                    text,
+                    is_final=False,
+                    consensus=False if consensus_model is not None else None,
+                )
         except Exception as e:
             emit_error(f"Transcription error: {e}")
 
     def transcribe_final(samples):
         nonlocal last_final_text
         if len(samples) < int(0.4 * SAMPLE_RATE):
+            emit_status("Whisper ignored very short utterance.")
             return
         samples = amplify(samples)
         try:
@@ -466,6 +471,7 @@ def main():
 
             text = _segment_filtered_text(result)
             if not text:
+                emit_status("Whisper final produced no accepted speech text.")
                 return
             if secondary is not None:
                 secondary_text = _segment_filtered_text(secondary)
@@ -483,13 +489,15 @@ def main():
             if text == last_final_text:
                 return
             last_final_text = text
-            # consensus=True only when BOTH models ran and converged — the
-            # downstream voicechat gate requires convergence for pass-through.
+            # consensus=True only when BOTH models ran and converged. When the
+            # second model is off/unavailable, omit consensus metadata entirely
+            # so legacy single-model consumers do not interpret consensus=false
+            # as an explicit dual-ASR rejection.
             emit_transcript(
                 text,
                 is_final=True,
                 doa=tuning.doa() if tuning else None,
-                consensus=consensus_model is not None,
+                consensus=True if consensus_model is not None else None,
             )
         except Exception as e:
             emit_error(f"Transcription error: {e}")

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.427",
+  "version": "1.0.429",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.427",
+      "version": "1.0.429",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.427",
+  "version": "1.0.429",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",