Merge pull request #3255 from LauraGPT/funasr-language-option

fix: allow configuring FunASR recognition language
This commit is contained in:
CGD
2026-06-30 14:57:31 +08:00
committed by GitHub
5 changed files with 108 additions and 1 deletions
@@ -0,0 +1,9 @@
-- Add language configuration for local FunASR ASR.
UPDATE `ai_model_provider`
SET `fields` = '[{"key":"model_dir","label":"模型目录","type":"string"},{"key":"output_dir","label":"输出目录","type":"string"},{"key":"language","label":"识别语言","type":"string","default":"auto"}]'
WHERE `id` = 'SYSTEM_ASR_FunASR';
UPDATE `ai_model_config`
SET `config_json` = JSON_SET(`config_json`, '$.language', 'auto')
WHERE `id` = 'ASR_FunASR'
AND JSON_EXTRACT(`config_json`, '$.language') IS NULL;
+7
View File
@@ -690,3 +690,10 @@ databaseChangeLog:
- sqlFile:
encoding: utf8
path: classpath:db/changelog/202606261131.sql
- changeSet:
id: 202606301405
author: LauraGPT
changes:
- sqlFile:
encoding: utf8
path: classpath:db/changelog/202606301405.sql
+3
View File
@@ -360,6 +360,9 @@ ASR:
type: fun_local
model_dir: models/SenseVoiceSmall
output_dir: tmp/
# 识别语种:auto 自动检测;如需限制只识别中文可设为 zh,避免中文短句误判为韩文等。
# SenseVoice 支持 zh、en、ja、ko、yue 等语种标记。
language: auto
FunASRServer:
# 独立部署FunASR,使用FunASR的API服务,只需要五句话
# 第一句:mkdir -p ./funasr-runtime-resources/models
@@ -50,6 +50,7 @@ class ASRProvider(ASRProviderBase):
self.interface_type = InterfaceType.LOCAL
self.model_dir = config.get("model_dir")
self.output_dir = config.get("output_dir") # 修正配置键名
self.language = config.get("language", "auto")
self.delete_audio_file = delete_audio_file
# 确保输出目录存在
@@ -80,7 +81,7 @@ class ASRProvider(ASRProviderBase):
self.model.generate,
input=artifacts.pcm_bytes,
cache={},
language="auto",
language=self.language,
use_itn=True,
batch_size_s=60,
)
@@ -0,0 +1,87 @@
import asyncio
import importlib
import sys
import types
from pathlib import Path
def install_fakes(monkeypatch):
class Logger:
def bind(self, **_kwargs):
return self
def debug(self, *_args, **_kwargs):
pass
def error(self, *_args, **_kwargs):
pass
def info(self, *_args, **_kwargs):
pass
def warning(self, *_args, **_kwargs):
pass
logger_module = types.ModuleType("config.logger")
logger_module.setup_logging = lambda: Logger()
config_module = types.ModuleType("config")
config_module.__path__ = []
monkeypatch.setitem(sys.modules, "config", config_module)
monkeypatch.setitem(sys.modules, "config.logger", logger_module)
psutil_module = types.ModuleType("psutil")
psutil_module.virtual_memory = lambda: types.SimpleNamespace(total=4 * 1024 * 1024 * 1024)
monkeypatch.setitem(sys.modules, "psutil", psutil_module)
class Base:
pass
base_module = types.ModuleType("core.providers.asr.base")
base_module.ASRProviderBase = Base
monkeypatch.setitem(sys.modules, "core.providers.asr.base", base_module)
dto_module = types.ModuleType("core.providers.asr.dto.dto")
dto_module.InterfaceType = types.SimpleNamespace(LOCAL="local")
monkeypatch.setitem(sys.modules, "core.providers.asr.dto.dto", dto_module)
utils_module = types.ModuleType("core.providers.asr.utils")
utils_module.lang_tag_filter = lambda text: {"content": text}
monkeypatch.setitem(sys.modules, "core.providers.asr.utils", utils_module)
class FakeAutoModel:
last_instance = None
def __init__(self, **_kwargs):
self.generate_kwargs = None
FakeAutoModel.last_instance = self
def generate(self, **kwargs):
self.generate_kwargs = kwargs
return [{"text": "你好"}]
funasr_module = types.ModuleType("funasr")
funasr_module.AutoModel = FakeAutoModel
monkeypatch.setitem(sys.modules, "funasr", funasr_module)
return FakeAutoModel
def test_fun_local_passes_configured_language_to_funasr_generate(tmp_path, monkeypatch):
server_root = Path(__file__).resolve().parents[1]
monkeypatch.syspath_prepend(str(server_root))
fake_model = install_fakes(monkeypatch)
monkeypatch.delitem(sys.modules, "core.providers.asr.fun_local", raising=False)
module = importlib.import_module("core.providers.asr.fun_local")
provider = module.ASRProvider(
{"model_dir": "models/SenseVoiceSmall", "output_dir": str(tmp_path), "language": "zh"},
delete_audio_file=False,
)
artifacts = types.SimpleNamespace(pcm_bytes=b"\0\0", file_path=str(tmp_path / "audio.wav"))
text, audio_path = asyncio.run(
provider.speech_to_text([], "session-1", audio_format="pcm", artifacts=artifacts)
)
assert text == {"content": "你好"}
assert audio_path == artifacts.file_path
assert fake_model.last_instance.generate_kwargs["language"] == "zh"