flightsimulatorplatfrom/OpenSquawk
1
0
Fork 0
mirror of https://github.com/OpenSquawk/OpenSquawk synced 2026-08-03 23:58:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat(atc): bias Whisper STT toward aviation phraseology

Browse Source 
Replace the generic one-line Whisper prompt with a vocabulary-rich bias prompt
(phonetic alphabet, aviation number words, common phraseology, and the airline
telephony names pulled live from DEFAULT_AIRLINE_TELEPHONY) and set
temperature: 0. Aviation R/T is a constrained domain, so seeding the expected
vocabulary improves transcription of callsigns, runways, and readbacks — the
upstream input to the decision engine. Stays within the ~224-token prompt cap.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
leubeem
2026-06-15 14:47:22 +02:00
parent dad224820b
commit 3485618468
1 changed files with 17 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
server/api/atc/ptt.post.ts
View File

@@ -11,9 +11,24 @@ import { TransmissionLog } from "../../models/TransmissionLog";
import { getUserFromEvent } from "../../utils/auth";
import { enforceRateLimit, getClientIp } from "../../utils/rateLimit";
import { recordUsage } from "../../utils/usage";
import { DEFAULT_AIRLINE_TELEPHONY } from "../../../shared/utils/radioSpeech";
type AudioFormat = 'wav' | 'mp3' | 'ogg' | 'webm'
// Whisper accepts a ~224-token `prompt` that biases recognition toward the
// expected vocabulary and spelling. Aviation R/T is a constrained domain, so
// seeding the phonetic alphabet, aviation number words, common phraseology, and
// the airline telephony names in play markedly improves transcription of
// callsigns, runways, and readbacks. The telephony names are pulled from the
// shared map so this stays in sync as airlines are added.
const STT_BIAS_PROMPT = [
"Air traffic control radio communication in ICAO English phraseology.",
"Phonetic alphabet: Alfa Bravo Charlie Delta Echo Foxtrot Golf Hotel India Juliett Kilo Lima Mike November Oscar Papa Quebec Romeo Sierra Tango Uniform Victor Whiskey X-ray Yankee Zulu.",
"Numbers: zero one two three four five six seven eight niner, also tree fife niner, decimal.",
`Airline callsigns: ${Object.values(DEFAULT_AIRLINE_TELEPHONY).join(", ")}.`,
"Common phrases: ready for pushback, request taxi, holding point, line up and wait, cleared for takeoff, contact tower, QNH, flight level, squawk, wilco, roger, affirm, negative, say again, runway, heading, descend, climb, maintain.",
].join(" ");
interface PTTRequest {
audio: string; // Base64 encoded audio
moduleId: string;
@@ -124,7 +139,8 @@ export default defineEventHandler(async (event) => {
file: createReadStream(audioFileForWhisper),
model: "whisper-1",
language: "en",
prompt: "This is ATC radio communication with aviation phraseology including callsigns, runway numbers, and standard procedures."
temperature: 0,
prompt: STT_BIAS_PROMPT
});
const transcribedText = transcription.text.trim();