fanlusky
a7779d2b05
`vendor/MetasequoiaImeEngine` submodule 改为 `engine/` 目录，导入的是 product-lock.json 原本钉住的提交 c810d201，逐文件核对过 blob，因此这是一次 搬迁而不是引擎升级。引擎自己的嵌套 submodule（googlepinyinime-rev、utfcpp、 miniaudio）按各自 gitlink 一并展开，来源与提交见 engine/UPSTREAM.md。 Windows 专门化裁掉了不属于本产品的部分：whisper.cpp 与本地推理 provider、 macOS 示例宿主、整个 dictionary/ 建库流水线、上游 CI 与 release-please 配置。 词库仍按 product-lock.json 在构建时取发布产物。 引擎不再进 product-lock.json：没有 gitlink 就没有可比对的东西，锁里写一个上游 commit 等于宣称这棵树仍等于上游，而第一个 Windows 专门化改动落地时这句话就假了。 verify-contracts / verify-checkout 随之移除，verify-published 保留。 CI 与工具链：引擎源码并入根 scripts/format.sh（删掉引擎自己的那份，避免两套排除 列表分叉）；引擎 voice.yml 里的 silero_vad.onnx 摘要校验移到本仓 lock job； engine-bump-triage 工作流与其测试删除；gitsubmodule 的 dependabot 频率由 daily 改为 monthly。 README、README.en、AGENTS.md、CONTRIBUTING、docs/product-release.md 及各子目录 文档同步说明新的位置与边界。
138 lines
7.3 KiB
C++
138 lines
7.3 KiB
C++
#include <msime/voice/cloud_stt_worker.h>
|
|
#include <msime/voice/text_polisher.h>
|
|
#include <msime/voice/vad.h>
|
|
#include <msime/voice/wav_writer.h>
|
|
#include <msime/voice/provider_protocol.h>
|
|
#include <cmath>
|
|
#include <future>
|
|
#include <iostream>
|
|
#include <limits>
|
|
#include <thread>
|
|
using namespace metasequoia::voice;
|
|
void require(bool ok, const char *message)
|
|
{
|
|
if (!ok)
|
|
throw std::runtime_error(message);
|
|
}
|
|
template <class F> void fails(F action)
|
|
{
|
|
bool failed = false;
|
|
try
|
|
{
|
|
action();
|
|
}
|
|
catch (const VoiceError &)
|
|
{
|
|
failed = true;
|
|
}
|
|
require(failed, "Expected VoiceError");
|
|
}
|
|
int main(int argc, char **argv)
|
|
{
|
|
try
|
|
{
|
|
if (argc == 1)
|
|
{
|
|
const std::string binary = std::string("RIFF\0binary", 11) + "----MetasequoiaVoice0";
|
|
const auto upload = make_transcription_request(binary, "fixture-model", "zh");
|
|
const auto boundary = upload.content_type.substr(upload.content_type.find("boundary=") + 9);
|
|
require(binary.find(boundary) == std::string::npos, "Multipart boundary collides with audio");
|
|
require(upload.body.find(binary) != std::string::npos, "Binary audio must retain NUL bytes");
|
|
require(upload.body.find("name=\"language\"\r\n\r\nzh") != std::string::npos, "Explicit recognition language");
|
|
require(make_transcription_request(binary, "model").body.find("name=\"language\"") == std::string::npos, "Omit language for providers that reject it");
|
|
fails([] { make_transcription_request({}, "model"); });
|
|
fails([] { make_transcription_request("audio", {}); });
|
|
fails([] { make_transcription_request(std::string(maximum_encoded_audio_bytes + 1, 'x'), "model"); });
|
|
for (const auto *response : {R"({"text":"水杉"})", R"({"transcription":"水杉"})", R"({"result":{"text":"水杉"}})"})
|
|
require(parse_transcription(response) == "水杉", "Platform transcription response variants");
|
|
for (const auto *response : {"invalid", "[]", R"({"text":42})", R"({"text":"","result":null})"})
|
|
fails([&] { parse_transcription(response); });
|
|
fails([] { parse_transcription(std::string(1024 * 1024 + 1, 'x')); });
|
|
require(parse_polished_text(R"({"choices":[{"message":{"content":"水杉"}}]})") == "水杉", "Polish response");
|
|
for (const auto *response : {"{}", R"({"choices":[]})", R"({"choices":[{"message":{"content":null}}]})", R"({"choices":[{"message":{"content":""}}]})"})
|
|
fails([&] { parse_polished_text(response); });
|
|
const auto polish_body = make_polish_request("model", "prompt", "quoted \"line\"\ntext");
|
|
require(polish_body.find("quoted \\\"line\\\"\\ntext") != std::string::npos, "Polish JSON escaping");
|
|
|
|
const auto wav = WavWriter::create_wav({0, 1, -1, 2});
|
|
require(wav.size() == 52 && wav[0] == 'R' && wav[24] == 0x80 && wav[25] == 0x3e, "WAV header");
|
|
require(wav[46] == 0xff && wav[47] == 0x7f && wav[48] == 1 && wav[49] == 0x80, "little-endian PCM");
|
|
fails([] { WavWriter::create_wav({std::numeric_limits<float>::quiet_NaN()}); });
|
|
fails([] { WavWriter::create_wav({0}, 48000); });
|
|
fails([] { WavWriter::create_wav(std::vector<float>(maximum_samples + 1)); });
|
|
const std::vector<float> longer_clip(maximum_samples + sample_rate / 2, 0.125f);
|
|
require(WavWriter::create_wav(longer_clip, sample_rate, longer_clip.size()).size() == longer_clip.size() * 2 + 44, "Host upload budget preserves padding and longer clips");
|
|
fails([] { WavWriter::create_wav({0, 0}, sample_rate, 1); });
|
|
VadSegmenter vad;
|
|
require(!vad.process(nullptr, 0), "Empty VAD block");
|
|
std::vector<float> silence(3200), speech(1600, 0.1f), tail(8000);
|
|
vad.process(silence.data(), silence.size());
|
|
vad.process(speech.data(), speech.size());
|
|
vad.process(tail.data(), tail.size());
|
|
require(vad.should_flush(), "Speech ending");
|
|
require(vad.take_audio().size() == 12800, "Pre-roll and silence retained");
|
|
require(vad.take_audio().empty() && !vad.should_flush(), "VAD reset");
|
|
vad.process(speech.data(), speech.size());
|
|
require(vad.take_audio().size() == speech.size(), "Second utterance isolated");
|
|
return 0;
|
|
}
|
|
const std::string base = argv[1];
|
|
auto options = [&](const std::string &path) { return RequestOptions{base + path, "fixture-model", "fixture-token", 3000, {}}; };
|
|
const std::vector<float> pcm(160, 0.125f);
|
|
std::cerr << "HTTP: transcription and error cases\n";
|
|
CloudSttWorker asr(options("/asr"));
|
|
require(asr.recognize(pcm) == "水杉 voice", "ASR UTF-8");
|
|
require(asr.recognize({}).empty(), "Empty audio needs no network");
|
|
for (const auto *path : {"/denied", "/invalid", "/missing", "/oversize", "/redirect"})
|
|
{
|
|
CloudSttWorker invalid(options(path));
|
|
fails([&] { invalid.recognize(pcm); });
|
|
}
|
|
std::cerr << "HTTP: timeout and cancellation\n";
|
|
auto timeout = options("/slow");
|
|
timeout.timeout_ms = 50;
|
|
CloudSttWorker slow(timeout);
|
|
fails([&] { slow.recognize(pcm); });
|
|
auto cancelled = options("/asr");
|
|
cancelled.cancelled = std::make_shared<std::atomic_bool>(true);
|
|
CloudSttWorker stopped(cancelled);
|
|
fails([&] { stopped.recognize(pcm); });
|
|
auto during = options("/slow");
|
|
during.cancelled = std::make_shared<std::atomic_bool>(false);
|
|
CloudSttWorker cancelling(during);
|
|
auto pending = std::async(std::launch::async, [&] { fails([&] { cancelling.recognize(pcm); }); });
|
|
std::this_thread::sleep_for(std::chrono::milliseconds(50));
|
|
during.cancelled->store(true);
|
|
pending.get();
|
|
std::cerr << "HTTP: polish and fallback\n";
|
|
TextPolisher polish(options("/polish"), "Clean transcription only");
|
|
require(polish.polish("嗯 水杉") == "水杉", "Polished text");
|
|
for (const auto *path : {"/denied", "/invalid", "/missing"})
|
|
{
|
|
TextPolisher fallback(options(path), "prompt");
|
|
require(fallback.polish("原始文本") == "原始文本", "Preserve original on error");
|
|
}
|
|
TextPolisher fallback(timeout, "prompt");
|
|
require(fallback.polish("原始文本") == "原始文本", "Preserve original on timeout");
|
|
// An independently destroyed provider must not tear down another provider's CURL runtime.
|
|
std::cerr << "HTTP: concurrent provider lifetimes\n";
|
|
std::vector<std::future<void>> workers;
|
|
for (int i = 0; i < 4; ++i)
|
|
workers.push_back(std::async(std::launch::async, [&] {
|
|
for (int n = 0; n < 5; ++n)
|
|
{
|
|
CloudSttWorker temporary(options("/asr"));
|
|
require(temporary.recognize(pcm) == "水杉 voice", "Independent provider lifetime");
|
|
}
|
|
}));
|
|
for (auto &worker : workers)
|
|
worker.get();
|
|
require(asr.recognize(pcm) == "水杉 voice", "Original provider still works");
|
|
}
|
|
catch (const std::exception &error)
|
|
{
|
|
std::cerr << error.what() << '\n';
|
|
return 1;
|
|
}
|
|
}
|