feat(stt): support multi-model STT with parallel execution and consensus NLU extraction

ai procurement/app/handlers/__init__.py

@@ -14,8 +14,8 @@ from app import drafts as draft_store
 from app.config import settings
 from app.nlu import understand
 from app.runtime import BotContext, LinkedUser
-from app.stt import transcribe
-from app.texts import CANCELLED, LINK_START, NOTHING_TO_COMMIT, STT_FAILED, UNKNOWN_HINT, VOICE_EMPTY
+from app.stt import transcribe, transcribe_multi
+from app.texts import CANCELLED, LINK_START, NOTHING_TO_COMMIT, STT_FAILED, UNKNOWN_HINT, VOICE_EMPTY, esc
 
 from . import commands, callbacks, quote, supplier, material, queries, decisions
 
@@ -40,6 +40,7 @@ async def handle_update(ctx: BotContext, update: dict) -> None:
 async def handle_message(ctx: BotContext, chat_id: int, message: dict) -> None:
     text = (message.get("text") or "").strip()
     source = "text"
+    voice_transcripts: list[dict] | None = None
 
     if text.startswith("/"):
         await commands.handle_command(ctx, chat_id, text)
@@ -51,17 +52,20 @@ async def handle_message(ctx: BotContext, chat_id: int, message: dict) -> None:
         file_id = (message.get("voice") or message.get("audio") or message.get("document"))["file_id"]
         try:
             audio = await ctx.bale.download_file(file_id)
-            text = await transcribe(settings, audio,
-                                    mime_type=(message.get("voice") or {}).get("mime_type"),
-                                    filename=(message.get("voice") or {}).get("file_name"))
+            voice_transcripts = await transcribe_multi(
+                settings,
+                audio,
+                mime_type=(message.get("voice") or {}).get("mime_type"),
+                filename=(message.get("voice") or {}).get("file_name"),
+            )
         except Exception as exc:
             detail = "هوش مصنوعی تنظیم نشده" if "not configured" in str(exc) else str(exc)[:200]
             await ctx.send(chat_id, STT_FAILED.format(detail=esc(detail)))
             return
-        if not text:
+        if not voice_transcripts:
             await ctx.send(chat_id, VOICE_EMPTY)
             return
-        await ctx._log(chat_id, "in", "voice", text)
+        text = voice_transcripts[0]["text"]
 
     # Pending multi-step interactions take the message first.
     flow = ctx.pending.get(chat_id)
@@ -91,9 +95,16 @@ async def handle_message(ctx: BotContext, chat_id: int, message: dict) -> None:
             return
 
     draft_summary = _draft_summary(target_draft) if target_draft else None
-    result = await understand(settings, text, draft_summary)
+    input_payload = voice_transcripts if voice_transcripts else text
+    result = await understand(settings, input_payload, draft_summary)
     intent, patch = result.get("intent"), result.get("patch") or {}
-    logger.info("chat %s intent=%s conf=%s (is_edit=%s)", chat_id, intent, result.get("confidence"), target_draft is not None)
+    consensus_text = result.get("consensus_text") or text
+
+    if source == "voice":
+        await ctx._log(chat_id, "in", "voice", consensus_text)
+
+    logger.info("chat %s intent=%s conf=%s consensus=%r (is_edit=%s)",
+                chat_id, intent, result.get("confidence"), consensus_text[:40], target_draft is not None)
 
     if intent in ("register_quote", "update_draft"):
         await quote.handle_entities(