omnius 1.0.427 → 1.0.428

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 ${this.model} model...`
         });
         const consensusEnv = String(process.env["OMNIUS_ASR_CONSENSUS"] ?? "").trim().toLowerCase();
-        const consensusModel = consensusEnv === "0" || consensusEnv === "off" || consensusEnv === "false" ? "off" : consensusEnv || (this.model === "tiny" ? "base" : "tiny");
+        const consensusModel = consensusEnv === "0" || consensusEnv === "off" || consensusEnv === "false" ? "off" : consensusEnv || defaultConsensusModel(this.model);
+        const consensusThreshold = String(process.env["OMNIUS_ASR_CONSENSUS_THRESHOLD"] ?? "0.45");
+        const silenceMs = String(process.env["OMNIUS_ASR_SILENCE_MS"] ?? "800");
         this.process = spawn29(
           pyPath,
           [
@@ -603585,7 +603596,11 @@ var init_listen = __esm({
             "--window-seconds",
             "10",
             "--consensus-model",
-            consensusModel
+            consensusModel,
+            "--consensus-threshold",
+            consensusThreshold,
+            "--silence-ms",
+            silenceMs
           ],
           {
             stdio: ["pipe", "pipe", "pipe"],
@@ -603625,9 +603640,13 @@ var init_listen = __esm({
                   });
                   break;
                 case "consensus_rejected":
-                  updateListenLiveState({
-                    lastStatus: `consensus rejected (${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();
@@ -604033,8 +604052,9 @@ 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;
@@ -697778,7 +697798,7 @@ 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 +697965,35 @@ ${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 +737160,11 @@ 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/i.test(msg)) {
+          writeContent(() => renderInfo(`[voicechat/asr] ${msg}`));
+        }
+      });
       const summaryRunner = {
         injectUserMessage(content) {
           if (activeTask?.runner) {

package/dist/scripts/live-whisper.py

@@ -483,13 +483,15 @@ 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": "1.0.428",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.427",
+      "version": "1.0.428",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.427",
+  "version": "1.0.428",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",