import { existsSync } from "node:fs";import { resolve } from "node:path";import { Effect, Schema } from "effect";import type { AppContext } from "../../app-context";import { resolveBinary, runCommandAsyncEffect } from "../../core/command";import { SttIntegrationError } from "../../services/stt";import type { SttMode } from "../../services/stt";import { TtsIntegrationError } from "../../services/tts";import type { TtsMode } from "../../services/tts";const AUDIO_DEFAULT_MODE = "strict";const AUDIO_TRANSCODE_TIMEOUT_MS = 60_000; export const parseField = (value: FormDataEntryValue | null): string | undefined => { if (typeof value !== "string") return undefined; const trimmed = value.trim(); return trimmed.length > 0 ? trimmed : undefined;}; export const parseMode = (value: FormDataEntryValue | null): SttMode => { const modeValue = (parseField(value) ?? AUDIO_DEFAULT_MODE).toLowerCase(); if (modeValue === "strict" || modeValue === "best_effort") { return modeValue; } throw new SttIntegrationError(400, "invalid_mode", "mode must be strict or best_effort");}; export const looksLikeWav = (bytes: Uint8Array): boolean => { if (bytes.length < 12) return false; const riff = String.fromCharCode(...bytes.slice(0, 4)); const wave = String.fromCharCode(...bytes.slice(8, 12)); return riff === "RIFF" && wave === "WAVE";}; type AudioModelError = new ( status: number, code: string, message: string, details?: Record<string, unknown>,) => Error; const resolveAudioModelPath = ( context: AppContext, requested: string | undefined, subdir: "stt" | "tts", envVariable: string, IntegrationError: AudioModelError,): { requestedModel: string; modelPath: string } => { const requestedModel = requested || process.env[envVariable]?.trim(); if (!requestedModel) { throw new IntegrationError( 400, "model_missing", `No ${subdir.toUpperCase()} model provided. Set model field or ${envVariable}.`, ); } const modelPath = requestedModel.includes("/") ? resolve(requestedModel) : resolve(context.config.models_dir, subdir, requestedModel); if (!existsSync(modelPath)) { throw new IntegrationError( 400, "model_not_found", `${subdir.toUpperCase()} model path does not exist`, { requested_model: requestedModel, resolved_model_path: modelPath }, ); } return { requestedModel, modelPath };}; export const resolveSttModelPath = ( context: AppContext, modelField: FormDataEntryValue | null,): { requestedModel: string; modelPath: string } => resolveAudioModelPath( context, parseField(modelField), "stt", "LOCAL_STUDIO_STT_MODEL", SttIntegrationError, ); export const resolveTtsModelPath = ( context: AppContext, modelValue: unknown,): { requestedModel: string; modelPath: string } => resolveAudioModelPath( context, typeof modelValue === "string" ? modelValue.trim() : undefined, "tts", "LOCAL_STUDIO_TTS_MODEL", TtsIntegrationError, ); export const ensureServiceLease = ( context: AppContext, mode: SttMode | TtsMode, serviceId: "stt" | "tts",): Effect.Effect<Record<string, unknown> | null, AudioDependencyError> => context.bridge.findInferenceProcess().pipe( Effect.mapError( (source) => new AudioDependencyError({ operation: "lease", message: `Could not inspect inference lease: ${String(source)}`, source, }), ), Effect.map((holder) => { if (!holder || mode === "best_effort") return null; return { code: "gpu_lease_conflict", requested_service: { id: serviceId }, holder_service: { id: "llm" }, actions: ["best_effort"], }; }), ); export class AudioDependencyError extends Schema.TaggedErrorClass<AudioDependencyError>()( "AudioDependencyError", { operation: Schema.Literals(["lease"]), message: Schema.String, source: Schema.Unknown, },) {} export const defaultTranscodeToWav = (options: { sourcePath: string; outputPath: string;}): Effect.Effect<string, SttIntegrationError> => Effect.gen(function* () { const ffmpegPath = resolveBinary(process.env["LOCAL_STUDIO_FFMPEG_CLI"] ?? "ffmpeg"); if (!ffmpegPath) { return yield* Effect.fail( new SttIntegrationError( 503, "ffmpeg_missing", "ffmpeg is required for non-WAV uploads. Install ffmpeg or upload WAV input.", ), ); } const result = yield* runCommandAsyncEffect( ffmpegPath, ["-y", "-i", options.sourcePath, "-ac", "1", "-ar", "16000", "-f", "wav", options.outputPath], { timeoutMs: AUDIO_TRANSCODE_TIMEOUT_MS }, ); if (result.timedOut) { return yield* Effect.fail( new SttIntegrationError(504, "audio_transcode_timeout", "Audio transcode timed out", { stderr: result.stderr, stdout: result.stdout, }), ); } if (result.status !== 0) { return yield* Effect.fail( new SttIntegrationError(400, "audio_transcode_failed", "Failed to transcode audio to WAV", { exit_code: result.status, signal: result.signal, stderr: result.stderr, stdout: result.stdout, }), ); } return options.outputPath; });
import { existsSync } from "node:fs";
import { resolve } from "node:path";
import { Effect, Schema } from "effect";
import type { AppContext } from "../../app-context";
import { resolveBinary, runCommandAsyncEffect } from "../../core/command";
import { SttIntegrationError } from "../../services/stt";
import type { SttMode } from "../../services/stt";
import { TtsIntegrationError } from "../../services/tts";
import type { TtsMode } from "../../services/tts";
const AUDIO_DEFAULT_MODE = "strict";
const AUDIO_TRANSCODE_TIMEOUT_MS = 60_000;
export const parseField = (value: FormDataEntryValue | null): string | undefined => {
if (typeof value !== "string") return undefined;
const trimmed = value.trim();
return trimmed.length > 0 ? trimmed : undefined;
};
export const parseMode = (value: FormDataEntryValue | null): SttMode => {
const modeValue = (parseField(value) ?? AUDIO_DEFAULT_MODE).toLowerCase();
if (modeValue === "strict" || modeValue === "best_effort") {
return modeValue;
}
throw new SttIntegrationError(400, "invalid_mode", "mode must be strict or best_effort");
};
export const looksLikeWav = (bytes: Uint8Array): boolean => {
if (bytes.length < 12) return false;
const riff = String.fromCharCode(...bytes.slice(0, 4));
const wave = String.fromCharCode(...bytes.slice(8, 12));
return riff === "RIFF" && wave === "WAVE";
};
type AudioModelError = new (
status: number,
code: string,
message: string,
details?: Record<string, unknown>,
) => Error;
const resolveAudioModelPath = (
context: AppContext,
requested: string | undefined,
subdir: "stt" | "tts",
envVariable: string,
IntegrationError: AudioModelError,
): { requestedModel: string; modelPath: string } => {
const requestedModel = requested || process.env[envVariable]?.trim();
if (!requestedModel) {
throw new IntegrationError(
400,
"model_missing",
`No ${subdir.toUpperCase()} model provided. Set model field or ${envVariable}.`,
);
}
const modelPath = requestedModel.includes("/")
? resolve(requestedModel)
: resolve(context.config.models_dir, subdir, requestedModel);
if (!existsSync(modelPath)) {
throw new IntegrationError(
400,
"model_not_found",
`${subdir.toUpperCase()} model path does not exist`,
{ requested_model: requestedModel, resolved_model_path: modelPath },
);
}
return { requestedModel, modelPath };
};
export const resolveSttModelPath = (
context: AppContext,
modelField: FormDataEntryValue | null,
): { requestedModel: string; modelPath: string } =>
resolveAudioModelPath(
context,
parseField(modelField),
"stt",
"LOCAL_STUDIO_STT_MODEL",
SttIntegrationError,
);
export const resolveTtsModelPath = (
context: AppContext,
modelValue: unknown,
): { requestedModel: string; modelPath: string } =>
resolveAudioModelPath(
context,
typeof modelValue === "string" ? modelValue.trim() : undefined,
"tts",
"LOCAL_STUDIO_TTS_MODEL",
TtsIntegrationError,
);
export const ensureServiceLease = (
context: AppContext,
mode: SttMode | TtsMode,
serviceId: "stt" | "tts",
): Effect.Effect<Record<string, unknown> | null, AudioDependencyError> =>
context.bridge.findInferenceProcess().pipe(
Effect.mapError(
(source) =>
new AudioDependencyError({
operation: "lease",
message: `Could not inspect inference lease: ${String(source)}`,
source,
}),
),
Effect.map((holder) => {
if (!holder || mode === "best_effort") return null;
return {
code: "gpu_lease_conflict",
requested_service: { id: serviceId },
holder_service: { id: "llm" },
actions: ["best_effort"],
};
}),
);
export class AudioDependencyError extends Schema.TaggedErrorClass<AudioDependencyError>()(
"AudioDependencyError",
{
operation: Schema.Literals(["lease"]),
message: Schema.String,
source: Schema.Unknown,
},
) {}
export const defaultTranscodeToWav = (options: {
sourcePath: string;
outputPath: string;
}): Effect.Effect<string, SttIntegrationError> =>
Effect.gen(function* () {
const ffmpegPath = resolveBinary(process.env["LOCAL_STUDIO_FFMPEG_CLI"] ?? "ffmpeg");
if (!ffmpegPath) {
return yield* Effect.fail(
new SttIntegrationError(
503,
"ffmpeg_missing",
"ffmpeg is required for non-WAV uploads. Install ffmpeg or upload WAV input.",
),
);
}
const result = yield* runCommandAsyncEffect(
ffmpegPath,
["-y", "-i", options.sourcePath, "-ac", "1", "-ar", "16000", "-f", "wav", options.outputPath],
{ timeoutMs: AUDIO_TRANSCODE_TIMEOUT_MS },
);
if (result.timedOut) {
return yield* Effect.fail(
new SttIntegrationError(504, "audio_transcode_timeout", "Audio transcode timed out", {
stderr: result.stderr,
stdout: result.stdout,
}),
);
}
if (result.status !== 0) {
return yield* Effect.fail(
new SttIntegrationError(400, "audio_transcode_failed", "Failed to transcode audio to WAV", {
exit_code: result.status,
signal: result.signal,
stderr: result.stderr,
stdout: result.stdout,
}),
);
}
return options.outputPath;
});