stt updates and intelligence updates

drb-c2-core/app/internal/intelligence.py

@@ -12,7 +12,7 @@ from typing import Optional
 from app.internal.logger import logger
 from app.internal import firestore as fstore
 
-_PROMPT_TEMPLATE = """You are analyzing a P25 public safety radio transcript. Extract structured information and respond ONLY with a single valid JSON object — no markdown, no explanation.
+_PROMPT_TEMPLATE = """You are analyzing a P25 public safety radio transcript. The audio was transcribed by Whisper through a digital radio vocoder, which introduces errors. Extract structured information and respond ONLY with a single valid JSON object — no markdown, no explanation.
 
 Schema:
 {{
@@ -21,7 +21,8 @@ Schema:
   "location": "most specific location string found, or empty string",
   "vehicles": [vehicle descriptions mentioned, e.g. "Hyundai Tucson", "black sedan"],
   "units": [unit IDs or officer numbers mentioned, e.g. "Unit 511", "Car 4"],
-  "severity": one of "minor" | "moderate" | "major" | "unknown"
+  "severity": one of "minor" | "moderate" | "major" | "unknown",
+  "transcript_corrected": "corrected transcript string, or null if no corrections needed"
 }}
 
 Rules:
@@ -29,7 +30,9 @@ Rules:
 - tags: be specific and lowercase, hyphenated. Do not repeat incident_type as a tag.
 - units: only identifiers explicitly mentioned, not inferred.
 - Do not invent details not present in the transcript.
+- transcript_corrected: fix only clear STT errors caused by vocoder distortion (e.g. "Several" → "10-4", misheard street names, garbled unit IDs). Keep all radio language as-is — do NOT decode codes into plain English. Return null if the transcript looks accurate.
 
+Talkgroup: {talkgroup_name}
 Transcript:
 {transcript}"""
 
@@ -37,17 +40,18 @@ Transcript:
 async def extract_tags(
     call_id: str,
     transcript: str,
+    talkgroup_name: Optional[str] = None,
 ) -> tuple[list[str], Optional[str], Optional[str]]:
     """
-    Extract incident tags, type, and location from a transcript via Gemini.
+    Extract incident tags, type, location, and corrected transcript via Gemini.
 
     Returns:
         (tags, primary_type, location)
 
     Side-effect: updates calls/{call_id} in Firestore with tags, location,
-    vehicles, units, severity; also stores the call embedding.
+    vehicles, units, severity, transcript_corrected; also stores the call embedding.
     """
-    result = await asyncio.to_thread(_sync_extract, transcript)
+    result = await asyncio.to_thread(_sync_extract, transcript, talkgroup_name)
 
     tags: list[str] = result.get("tags") or []
     incident_type: Optional[str] = result.get("incident_type") or None
@@ -55,6 +59,7 @@ async def extract_tags(
     vehicles: list[str] = result.get("vehicles") or []
     units: list[str] = result.get("units") or []
     severity: str = result.get("severity") or "unknown"
+    transcript_corrected: Optional[str] = result.get("transcript_corrected") or None
 
     if incident_type in ("unknown", "other", ""):
         incident_type = None
@@ -74,6 +79,8 @@ async def extract_tags(
         updates["units"] = units
     if embedding:
         updates["embedding"] = embedding
+    if transcript_corrected:
+        updates["transcript_corrected"] = transcript_corrected
 
     try:
         await fstore.doc_set("calls", call_id, updates)
@@ -82,12 +89,13 @@ async def extract_tags(
 
     logger.info(
         f"Intelligence: call {call_id} → type={incident_type}, "
-        f"tags={tags}, location={location!r}, severity={severity}"
+        f"tags={tags}, location={location!r}, severity={severity}, "
+        f"corrected={transcript_corrected is not None}"
     )
     return tags, incident_type, location
 
 
-def _sync_extract(transcript: str) -> dict:
+def _sync_extract(transcript: str, talkgroup_name: Optional[str]) -> dict:
     """Call Gemini Flash and parse the JSON response."""
     from app.config import settings
     import google.generativeai as genai
@@ -102,8 +110,13 @@ def _sync_extract(transcript: str) -> dict:
         generation_config={"response_mime_type": "application/json"},
     )
 
+    prompt = _PROMPT_TEMPLATE.format(
+        transcript=transcript,
+        talkgroup_name=talkgroup_name or "unknown",
+    )
+
     try:
-        response = model.generate_content(_PROMPT_TEMPLATE.format(transcript=transcript))
+        response = model.generate_content(prompt)
         return json.loads(response.text)
     except json.JSONDecodeError as e:
         logger.warning(f"Gemini returned non-JSON: {e}")

drb-c2-core/app/internal/transcription.py

@@ -11,17 +11,34 @@ from typing import Optional
 from app.internal.logger import logger
 from app.internal import firestore as fstore
 
+# Whisper treats `prompt` as preceding transcript text, not instructions.
+# Writing it as actual radio speech primes the vocabulary toward P25 codes
+# and phrasing before the model hears the audio.
+_WHISPER_PROMPT = (
+    "10-4. 10-23. 10-20. 10-97. 10-8. 10-7. 10-34. 10-50. 10-52. "
+    "Post 4, I'm out. Post 3. En route. On scene. In route. "
+    "Copy. Negative. Stand by. Be advised. Go ahead. "
+    "Units responding. Dispatch. Talkgroup. "
+    "Engine. Ladder. Medic. Rescue. Car. Unit. "
+    "MVA. MVC. Structure fire. Working fire."
+)
 
-async def transcribe_call(call_id: str, gcs_uri: str) -> Optional[str]:
+
+async def transcribe_call(
+    call_id: str,
+    gcs_uri: str,
+    talkgroup_name: Optional[str] = None,
+) -> Optional[str]:
     """
     Transcribe audio at the given GCS URI and store the result in Firestore.
 
     Args:
-        call_id: Firestore document ID in the 'calls' collection.
-        gcs_uri:  GCS URI of the audio file, e.g. gs://bucket/calls/xyz.mp3
+        call_id:        Firestore document ID in the 'calls' collection.
+        gcs_uri:        GCS URI of the audio file, e.g. gs://bucket/calls/xyz.mp3
+        talkgroup_name: Passed through to the intelligence layer; unused here.
 
     Returns:
-        The transcript string, or None if transcription failed / was skipped.
+        The raw Whisper transcript string, or None if transcription failed.
     """
     if not gcs_uri or not gcs_uri.startswith("gs://"):
         return None
@@ -53,11 +70,9 @@ def _sync_transcribe(gcs_uri: str) -> Optional[str]:
         logger.warning("OPENAI_API_KEY not set — transcription disabled.")
         return None
 
-    # Parse gs://bucket/path/to/file.mp3
     without_scheme = gcs_uri[len("gs://"):]
     bucket_name, blob_path = without_scheme.split("/", 1)
 
-    # Download to a temp file
     if settings.gcp_credentials_path:
         creds = service_account.Credentials.from_service_account_file(
             settings.gcp_credentials_path,
@@ -83,7 +98,7 @@ def _sync_transcribe(gcs_uri: str) -> Optional[str]:
                 model="whisper-1",
                 file=f,
                 language="en",
-                prompt="Public safety radio communication. May include police codes, fire, EMS, talkgroup IDs, unit numbers, addresses.",
+                prompt=_WHISPER_PROMPT,
             )
         return response.text.strip() or None
     finally: