mirror of
https://github.com/stablyai/orca.git
synced 2026-11-13 09:27:29 +00:00
Kelvin Amoaba
0b8e94aa5c
* feat(speech): run GGUF models on transcribe-cpp, falling back to the CPU The speech host child can now load a second engine. A model whose files are GGUF runs on transcribe-cpp (Metal on Apple silicon, Vulkan on Windows and Linux, CPU otherwise); everything else stays on sherpa-onnx, which remains the only engine on Intel Macs. A GPU that fails to load, errors, stalls for 25 s or crashes the process moves the dictation to a CPU-only process (TRANSCRIBE_BACKENDS=cpu) that replays the audio the GPU had not decoded yet, so no words are lost or repeated. A GPU that crashed or would not load is remembered per transcribe-cpp version and OS release, and voice.speechGpu = 'off' keeps it on the CPU. Packaging copies transcribe-cpp, koffi and the target's native package into resources/node_modules outside asar, fails when a target's package is missing, and ships the MSVC runtime beside transcribe.dll on Windows. No catalog model is GGUF yet, so nothing runs on transcribe-cpp until the catalog change that follows. * fix(speech): keep the speech queue running after a failed task, and drop audio frames after a discard * fix(speech): keep the engine's file reads off the main thread's hot paths The engine lookup and the GPU failure marker are read once per process instead of on every dictation start and mobile catalog request, the marker is written asynchronously, and the child reads the hotwords file with fs/promises. The baseline moves the existing engine presence checks from stt-worker-paths to the shared package table and adds the once-per-process marker reads.
98 lines
4.0 KiB
JavaScript
98 lines
4.0 KiB
JavaScript
const { cpSync, existsSync } = require('node:fs')
|
|
const { execFileSync } = require('node:child_process')
|
|
const { join } = require('node:path')
|
|
|
|
const TRANSCRIBE_CPP_PACKAGES = require('../src/shared/transcribe-cpp-packages.json')
|
|
// transcribe.dll needs MSVC runtime >= 14.40 (older ones crash in std::mutex); the system copy may be older.
|
|
const WINDOWS_CRT_DLLS = ['msvcp140.dll', 'vcruntime140.dll', 'vcruntime140_1.dll']
|
|
const MIN_WINDOWS_CRT_VERSION = [14, 40]
|
|
|
|
/**
|
|
* The packages the speech host loads transcribe-cpp from, in load order: the ESM binding, koffi
|
|
* (its FFI), koffi's prebuilt addon, and the target's native libraries. Each resolves from the
|
|
* previous one's directory, as Node does at runtime. Null where the target ships no transcribe-cpp.
|
|
*/
|
|
function transcribeCppRuntimePackages(readPackage, electronPlatformName, architecture) {
|
|
const native = TRANSCRIBE_CPP_PACKAGES[`${electronPlatformName}-${architecture}`]
|
|
if (!native) {
|
|
return null
|
|
}
|
|
const binding = readPackage('transcribe-cpp')
|
|
const koffi = readPackage('koffi', binding.packageDir)
|
|
const koffiAddon = readPackage(
|
|
`@koromix/koffi-${electronPlatformName}-${architecture}`,
|
|
koffi.packageDir
|
|
)
|
|
const nativeLibraries = readPackage(native, binding.packageDir)
|
|
return [binding, koffi, koffiAddon, nativeLibraries]
|
|
}
|
|
|
|
function readFileVersion(dll) {
|
|
const version = execFileSync(
|
|
'powershell.exe',
|
|
['-NoProfile', '-Command', `(Get-Item -LiteralPath '${dll}').VersionInfo.FileVersion`],
|
|
{ encoding: 'utf8' }
|
|
).trim()
|
|
return version.split(/[.,\s]+/).map((part) => Number.parseInt(part, 10))
|
|
}
|
|
|
|
function copyWindowsCrt(nativeDir) {
|
|
const systemDir = join(process.env.SystemRoot || 'C:\\Windows', 'System32')
|
|
for (const name of WINDOWS_CRT_DLLS) {
|
|
const source = join(systemDir, name)
|
|
if (!existsSync(source)) {
|
|
throw new Error(
|
|
`Packaging transcribe-cpp needs ${source} (Visual C++ 2015-2022 x64 runtime).`
|
|
)
|
|
}
|
|
const [major, minor] = readFileVersion(source)
|
|
if (!Number.isFinite(major) || !Number.isFinite(minor)) {
|
|
throw new Error(`Cannot read the file version of ${source}.`)
|
|
}
|
|
const [minMajor, minMinor] = MIN_WINDOWS_CRT_VERSION
|
|
if (major < minMajor || (major === minMajor && minor < minMinor)) {
|
|
throw new Error(
|
|
`${source} is ${major}.${minor}; transcribe-cpp needs ${MIN_WINDOWS_CRT_VERSION.join('.')}+. ` +
|
|
'Install the latest Visual C++ 2015-2022 x64 redistributable on the build host.'
|
|
)
|
|
}
|
|
// App-local beside transcribe.dll, which loads its dependencies from its own directory.
|
|
cpSync(source, join(nativeDir, name))
|
|
}
|
|
}
|
|
|
|
/** Copy transcribe-cpp into the slice's resources/node_modules, outside asar where koffi can dlopen it. */
|
|
function copyPackagedTranscribeCpp(readPackage, resourcesDir, electronPlatformName, architecture) {
|
|
// beforePack's assertPackagedNativeVariantsInstalled already failed any real build missing these;
|
|
// only an afterPack driven without it (packaging tests) gets here without them.
|
|
if (missingTranscribeCppPackages(readPackage, electronPlatformName, architecture).length > 0) {
|
|
return
|
|
}
|
|
const packages = transcribeCppRuntimePackages(readPackage, electronPlatformName, architecture)
|
|
if (!packages) {
|
|
return
|
|
}
|
|
for (const { name, packageDir } of packages) {
|
|
const target = join(resourcesDir, 'node_modules', ...name.split('/'))
|
|
cpSync(packageDir, target, { recursive: true, dereference: true })
|
|
}
|
|
if (electronPlatformName === 'win32') {
|
|
copyWindowsCrt(join(resourcesDir, 'node_modules', ...packages[3].name.split('/')))
|
|
}
|
|
}
|
|
|
|
/** The transcribe-cpp packages a target needs that this install lacks (install:release adds other CPUs). */
|
|
function missingTranscribeCppPackages(readPackage, electronPlatformName, architecture) {
|
|
try {
|
|
transcribeCppRuntimePackages(readPackage, electronPlatformName, architecture)
|
|
return []
|
|
} catch (error) {
|
|
return [String(error.message ?? error)]
|
|
}
|
|
}
|
|
|
|
module.exports = {
|
|
copyPackagedTranscribeCpp,
|
|
missingTranscribeCppPackages
|
|
}
|