Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

docs(site): add SenseVoice native server deployment

LauraGPT2469fb73e76e
diff --git a/web-pages/product-site/data/deployments.json b/web-pages/product-site/data/deployments.json
index e09bbd96a..c64066f52 100644
--- a/web-pages/product-site/data/deployments.json
+++ b/web-pages/product-site/data/deployments.json
@@ -236,6 +236,95 @@
}
}
},
+ {
+ "id": "sensevoice-native-server",
+ "routes": {"zh": "/deploy/sensevoice-native-server.html", "en": "/en/deploy/sensevoice-native-server.html"},
+ "maturity": "production-verified",
+ "selector_rank": 20.5,
+ "workloads": ["private-api", "realtime", "edge"],
+ "hardware": ["cpu", "nvidia-gpu"],
+ "priorities": ["compatibility", "portability"],
+ "models": ["SenseVoiceSmall-GGUF", "FSMN-VAD-GGUF"],
+ "operating_systems": ["Linux"],
+ "interfaces": ["OpenAI-compatible HTTP", "OpenAI realtime WebSocket", "WS /v1/realtime?intent=transcription", "SSE", "SRT/VTT"],
+ "tested": {
+ "funasr": "SenseVoice main@b054623c",
+ "runtime": "sensevoice-server@558bd67c",
+ "verified": "2026-08-13"
+ },
+ "commands": {
+ "install": [
+ "git clone https://github.com/QwenAudio/SenseVoice.git && cd SenseVoice",
+ "git checkout b054623cca8f015b73ec471dce4f473ac47413da",
+ "python3 -m pip install -U huggingface_hub",
+ "bash runtime/llama.cpp/download-funasr-model.sh sensevoice model",
+ "cmake -S runtime/llama.cpp -B runtime/llama.cpp/build -DCMAKE_BUILD_TYPE=Release",
+ "cmake --build runtime/llama.cpp/build --target sensevoice-server -j"
+ ],
+ "launch": [
+ "runtime/llama.cpp/build/bin/sensevoice-server -m model/sensevoice-small-q8.gguf -vad model/fsmn-vad.gguf --threads 8 --max-connections 4 --max-audio-seconds 300 127.0.0.1 8040"
+ ],
+ "health": [
+ "curl -fsS http://127.0.0.1:8040/health",
+ "curl -fsS http://127.0.0.1:8040/v1/models"
+ ],
+ "smoke": [
+ "curl -fsS http://127.0.0.1:8040/v1/audio/transcriptions -F [email protected]",
+ "curl -fsS http://127.0.0.1:8040/v1/audio/transcriptions -F [email protected] -F response_format=vtt | tee transcript.vtt && grep -q WEBVTT transcript.vtt",
+ "curl -fsSN http://127.0.0.1:8040/v1/audio/transcriptions -F [email protected] -F stream=true | grep -q transcript.text.done",
+ "python3 runtime/llama.cpp/tests/stream_client.py 127.0.0.1 8040 speech_16k.wav 200"
+ ]
+ },
+ "evidence": [
+ {
+ "label": "merged SenseVoice server contribution",
+ "url": "https://github.com/QwenAudio/SenseVoice/pull/341"
+ },
+ {
+ "label": "source and operating guide at the verified merge",
+ "url": "https://github.com/QwenAudio/SenseVoice/blob/b054623cca8f015b73ec471dce4f473ac47413da/runtime/llama.cpp/sensevoice-server/README.md"
+ },
+ {
+ "label": "pull-request build, unit, and contract workflow",
+ "url": "https://github.com/QwenAudio/SenseVoice/actions/runs/31633373807"
+ },
+ {
+ "label": "main-branch build, unit, and contract workflow",
+ "url": "https://github.com/QwenAudio/SenseVoice/actions/runs/31633730096"
+ },
+ {
+ "label": "server regression and contract tests",
+ "url": "https://github.com/QwenAudio/SenseVoice/tree/b054623cca8f015b73ec471dce4f473ac47413da/runtime/llama.cpp/tests"
+ }
+ ],
+ "benchmarks": [],
+ "translations": {
+ "zh": {
+ "name": "SenseVoice 原生 OpenAI 实时服务",
+ "summary": "用一个无 Python 运行时的 C++ 二进制提供 SenseVoice REST、SSE、SRT/VTT 与 OpenAI Realtime WebSocket 转写。",
+ "fit": ["本地或内网 OpenAI 兼容语音接口", "CPU 优先、可选 CUDA 的低依赖部署", "需要文件转写、流式 partial/final 与服务端 VAD"],
+ "not_fit": ["需要预编译跨平台发布包", "需要多租户认证、计费或持久任务队列", "没有在目标硬件完成容量测试就直接暴露公网"],
+ "selection_reason": "原生 GGUF 服务把 REST 文件转写和实时 WebSocket 放进同一个受资源上限保护的二进制,适合私有兼容接口和边缘流式场景。",
+ "primary_limitation": "当前官方验证覆盖 Linux 源码构建;服务不内置生产认证、TLS、多租户配额或持久队列,必须由网关和基础设施补齐。",
+ "status_label": "生产验证",
+ "operations": ["固定 merge commit、SenseVoice GGUF 与 FSMN-VAD GGUF", "显式设置连接上限、单请求音频上限和读取超时", "发布前分别压测 REST、SSE 和 WebSocket,并保留旧二进制回滚"],
+ "security": ["默认绑定 127.0.0.1 或可信内网", "在反向代理实施 TLS、认证、MIME、请求体大小、速率和并发限制", "只加载经过来源与摘要校验的 GGUF"],
+ "troubleshooting": ["先用 CPU、已知 16 kHz WAV 和 REST JSON 路径验证", "确认模型文件名为 sensevoice-small-q8.gguf 与 fsmn-vad.gguf", "WebSocket 客户端发送 16 kHz 单声道 PCM16 的 base64 chunk"]
+ },
+ "en": {
+ "name": "SenseVoice native OpenAI realtime server",
+ "summary": "Use one C++ binary with no Python runtime to serve SenseVoice over REST, SSE, SRT/VTT, and OpenAI realtime WebSocket transcription.",
+ "fit": ["A local or private OpenAI-compatible speech endpoint", "CPU-first deployment with optional CUDA and few runtime dependencies", "File transcription, streaming partial/final results, and server-side VAD"],
+ "not_fit": ["Prebuilt cross-platform release packages are required", "Built-in multi-tenant authentication, billing, or durable job queues are required", "Public exposure before capacity testing on the target hardware"],
+ "selection_reason": "The native GGUF server combines REST file transcription and realtime WebSocket in one resource-bounded binary for private compatible APIs and edge streaming.",
+ "primary_limitation": "Official verification currently covers Linux source builds; the server does not include production authentication, TLS, tenant quotas, or durable queues, so provide them at the gateway and infrastructure layers.",
+ "status_label": "Production verified",
+ "operations": ["Pin the merge commit plus the SenseVoice and FSMN-VAD GGUF files", "Set the connection limit, per-request audio limit, and read timeout explicitly", "Load-test REST, SSE, and WebSocket separately before rollout, and retain the previous binary for rollback"],
+ "security": ["Bind to 127.0.0.1 or a trusted private network by default", "Enforce TLS, authentication, MIME, request-body size, rate, and concurrency at the reverse proxy", "Load only GGUF files with verified provenance and digests"],
+ "troubleshooting": ["Validate CPU, a known 16 kHz WAV, and the REST JSON path first", "Confirm the model files are named sensevoice-small-q8.gguf and fsmn-vad.gguf", "Send base64 chunks of 16 kHz mono PCM16 from WebSocket clients"]
+ }
+ }
+ },
{
"id": "audio-cpp",
"routes": {"zh": "/deploy/audio-cpp.html", "en": "/en/deploy/audio-cpp.html"},
diff --git a/web-pages/product-site/tests/test_output.py b/web-pages/product-site/tests/test_output.py
index 2c3af8357..81d2d9146 100644
--- a/web-pages/product-site/tests/test_output.py
+++ b/web-pages/product-site/tests/test_output.py
@@ -77,6 +77,30 @@ def test_llama_cpp_pages_render_v020_download_matrix(built_site, relative, bound
assert boundary in soup.get_text(' ', strip=True)
+ ('relative', 'language_marker'),
+ (
+ ('deploy/sensevoice-native-server.html', '连接上限'),
+ ('en/deploy/sensevoice-native-server.html', 'connection limit'),
+ ),
+)
+def test_sensevoice_native_server_pages_render_operational_contract(
+ built_site, relative, language_marker
+):
+ soup = read_soup(built_site / relative)
+ text = soup.get_text(' ', strip=True)
+
+ for marker in (
+ 'sensevoice-server',
+ '/v1/audio/transcriptions',
+ '/v1/realtime?intent=transcription',
+ '--max-connections',
+ '--max-audio-seconds',
+ language_marker,
+ ):
+ assert marker in text
+
+
def test_benchmark_rows_have_complete_conditions(built_site):
registry = load_registry(SITE_ROOT / 'data' / 'deployments.json')
records = [record for entry in registry['deployments'] for record in entry['benchmarks']]
diff --git a/web-pages/product-site/tests/test_registry.py b/web-pages/product-site/tests/test_registry.py
index ff19165a6..720a6e1a0 100644
--- a/web-pages/product-site/tests/test_registry.py
+++ b/web-pages/product-site/tests/test_registry.py
@@ -18,6 +18,7 @@ EXPECTED_IDS = {
'vllm',
'sensevoice-tensorrt',
'llama-cpp',
+ 'sensevoice-native-server',
'audio-cpp',
'openai-api',
'realtime',
@@ -185,6 +186,54 @@ def test_llama_cpp_contract_tracks_v020_release_assets(valid_registry):
assert 'AMD' in entry['translations']['en']['primary_limitation']
+def test_sensevoice_native_server_contract_tracks_merged_runtime(valid_registry):
+ entry = next(
+ item for item in valid_registry['deployments']
+ if item['id'] == 'sensevoice-native-server'
+ )
+ llama_cpp = next(item for item in valid_registry['deployments'] if item['id'] == 'llama-cpp')
+ audio_cpp = next(item for item in valid_registry['deployments'] if item['id'] == 'audio-cpp')
+
+ assert valid_registry['verified'] == '2026-08-13'
+ assert entry['maturity'] == 'production-verified'
+ assert llama_cpp['selector_rank'] < entry['selector_rank'] < audio_cpp['selector_rank']
+ assert entry['tested'] == {
+ 'funasr': 'SenseVoice main@b054623c',
+ 'runtime': 'sensevoice-server@558bd67c',
+ 'verified': '2026-08-13',
+ }
+ assert entry['models'] == ['SenseVoiceSmall-GGUF', 'FSMN-VAD-GGUF']
+ assert entry['operating_systems'] == ['Linux']
+ assert {
+ 'OpenAI-compatible HTTP',
+ 'OpenAI realtime WebSocket',
+ 'SSE',
+ 'SRT/VTT',
+ } <= set(entry['interfaces'])
+ assert any(
+ 'git checkout b054623cca8f015b73ec471dce4f473ac47413da' in command
+ for command in entry['commands']['install']
+ )
+ assert any(
+ 'download-funasr-model.sh sensevoice' in command
+ for command in entry['commands']['install']
+ )
+ launch = '\n'.join(entry['commands']['launch'])
+ for marker in ('sensevoice-server', '--max-connections', '--max-audio-seconds'):
+ assert marker in launch
+ smoke = '\n'.join(entry['commands']['smoke'])
+ for marker in ('/v1/audio/transcriptions', 'response_format=vtt', 'stream_client.py'):
+ assert marker in smoke
+ assert any('/health' in command for command in entry['commands']['health'])
+ assert any('/v1/models' in command for command in entry['commands']['health'])
+ evidence_urls = {item['url'] for item in entry['evidence']}
+ assert 'https://github.com/QwenAudio/SenseVoice/pull/341' in evidence_urls
+ assert 'https://github.com/QwenAudio/SenseVoice/actions/runs/31633730096' in evidence_urls
+ limitation = entry['translations']['en']['primary_limitation'].lower()
+ assert 'linux' in limitation
+ assert 'authentication' in limitation
+
+
def test_download_assets_require_https_and_sha256(valid_registry):
data = copy.deepcopy(valid_registry)
entry = next(item for item in data['deployments'] if item['id'] == 'llama-cpp')
diff --git a/web-pages/product-site/tests/test_selector.py b/web-pages/product-site/tests/test_selector.py
index 30f1ee0bf..3dfa0821f 100644
--- a/web-pages/product-site/tests/test_selector.py
+++ b/web-pages/product-site/tests/test_selector.py
@@ -23,6 +23,7 @@ def entries():
[
('batch', 'nvidia-gpu', 'throughput', 'vllm'),
('edge', 'cpu', 'portability', 'llama-cpp'),
+ ('realtime', 'cpu', 'compatibility', 'sensevoice-native-server'),
('private-api', 'kubernetes', 'compatibility', 'containers'),
('realtime', 'nvidia-gpu', 'latency', 'realtime'),
],