flightsimulatorplatfrom/OpenSquawk
1
0
Fork 0
mirror of https://github.com/OpenSquawk/OpenSquawk synced 2026-08-01 14:16:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat(tts): map logical pool voices to distinct Piper speakers on Speaches

Browse Source 
Until now the Speaches branch sent every request to the single env-configured
model (de_DE-thorsten — a German voice for English radio), so the voice pool
had no audible effect. A server-side registry now resolves the logical voice
ids to disjoint controller/pilot Piper speakers; unknown ids keep the env
fallback and the cache key follows the resolved pair.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
itsrubberduck
2026-07-19 10:45:49 +02:00
parent b87884a608
commit b359670d47
3 changed files with 95 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
server/api/atc/say.post.ts
View File

@@ -10,6 +10,7 @@ import { TransmissionLog } from "../../models/TransmissionLog";
import { requireUserSession } from "../../utils/auth";
import { enforceRateLimit } from "../../utils/rateLimit";
import { recordUsage } from "../../utils/usage";
import { resolveSpeachesVoice } from "../../utils/voiceRegistry";
function outDir() {
@@ -204,8 +205,18 @@ export default defineEventHandler(async (event) => {
const useSpeaches = runtimeConfig.useSpeaches;
const usePiper = !useSpeaches && runtimeConfig.usePiper;
const provider = resolveTtsProvider(useSpeaches, usePiper);
const providerModel = provider === "speaches"
? (runtimeConfig.speechModelId || "speaches-ai/piper-en_US-ryan-low")
// Speaches: logical pool voices (alloy, verse, …) resolve to Piper
// model+voice pairs; anything unknown keeps the env-configured pair.
const speachesFallback = {
model: runtimeConfig.speechModelId || "speaches-ai/piper-en_US-ryan-low",
voice,
};
const speachesVoice = provider === "speaches"
? resolveSpeachesVoice(voice, speachesFallback)
: null;
const ttsVoice = speachesVoice?.voice ?? voice;
const providerModel = speachesVoice
? speachesVoice.model
: provider === "piper"
? "piper-local"
: TTS_MODEL;
@@ -221,7 +232,7 @@ export default defineEventHandler(async (event) => {
? buildFlightLabCacheKey({
normalized,
level,
voice,
voice: ttsVoice,
speed,
format: fmt,
provider,
@@ -263,11 +274,11 @@ export default defineEventHandler(async (event) => {
if (!audioBuffer && useSpeaches) {
// Speaches (prefer compact: MP3, otherwise FLAC/WAV/PCM)
const baseUrl = runtimeConfig.speachesBaseUrl || "";
const model = runtimeConfig.speechModelId || "speaches-ai/piper-en_US-ryan-low";
const model = providerModel;
if (!baseUrl) {
throw new Error("SPEACHES_BASE_URL not set");
}
audioBuffer = await speachesTTS(normalized, voice, model, fmt, baseUrl, speed);
audioBuffer = await speachesTTS(normalized, ttsVoice, model, fmt, baseUrl, speed);
modelUsed = model;
// Server returns the correct format according to response_format
actualMime = fmtToMime(fmt);