update config of language

Former-commit-id: 6c237bee605ec8fe2e28517a2d7e0a385747ab5b

pyproject.toml

@@ -7,9 +7,11 @@ requires-python = ">=3.11"
 dependencies = [
     "av>=14.2.0",
     "librosa>=0.11.0",
+    "llama-cpp-python>=0.3.8",
     "numpy>=2.1.3",
     "onnxruntime>=1.21.0",
     "pyaudio>=0.2.14",
+    "pywhispercpp>=1.3.0",
     "setuptools>=78.1.0",
     "soundfile>=0.13.1",
     "torch>=2.6.0",

requirements.txt

@@ -14,6 +14,8 @@ coloredlogs==15.0.1
     # via onnxruntime
 decorator==5.2.1
     # via librosa
+diskcache==5.6.3
+    # via llama-cpp-python
 filelock==3.18.0
     # via torch
 flatbuffers==25.2.10
@@ -25,7 +27,9 @@ humanfriendly==10.0
 idna==3.10
     # via requests
 jinja2==3.1.6
-    # via torch
+    # via
+    #   llama-cpp-python
+    #   torch
 joblib==1.4.2
     # via
     #   librosa
@@ -34,6 +38,8 @@ lazy-loader==0.4
     # via librosa
 librosa==0.11.0
     # via trans (pyproject.toml)
+llama-cpp-python==0.3.8
+    # via trans (pyproject.toml)
 llvmlite==0.44.0
     # via numba
 markupsafe==3.0.2
@@ -50,8 +56,10 @@ numpy==2.1.3
     # via
     #   trans (pyproject.toml)
     #   librosa
+    #   llama-cpp-python
     #   numba
     #   onnxruntime
+    #   pywhispercpp
     #   scikit-learn
     #   scipy
     #   soundfile
@@ -64,7 +72,9 @@ packaging==24.2
     #   onnxruntime
     #   pooch
 platformdirs==4.3.7
-    # via pooch
+    # via
+    #   pooch
+    #   pywhispercpp
 pooch==1.8.2
     # via librosa
 protobuf==6.30.2
@@ -73,8 +83,12 @@ pyaudio==0.2.14
     # via trans (pyproject.toml)
 pycparser==2.22
     # via cffi
+pywhispercpp==1.3.0
+    # via trans (pyproject.toml)
 requests==2.32.3
-    # via pooch
+    # via
+    #   pooch
+    #   pywhispercpp
 scikit-learn==1.6.1
     # via librosa
 scipy==1.15.2
@@ -98,10 +112,13 @@ threadpoolctl==3.6.0
 torch==2.6.0
     # via trans (pyproject.toml)
 tqdm==4.67.1
-    # via trans (pyproject.toml)
+    # via
+    #   trans (pyproject.toml)
+    #   pywhispercpp
 typing-extensions==4.13.1
     # via
     #   librosa
+    #   llama-cpp-python
     #   torch
 urllib3==2.3.0
     # via requests
@@ -109,6 +126,3 @@ websocket-client==1.8.0
     # via trans (pyproject.toml)
 websockets==15.0.1
     # via trans (pyproject.toml)
-
-llama-cpp-python
-pywhispercpp

run_server.py

@@ -22,7 +22,7 @@ if __name__ == "__main__":
     if "OMP_NUM_THREADS" not in os.environ:
         os.environ["OMP_NUM_THREADS"] = str(args.omp_num_threads)
 
-    from transcribe.server import TranscriptionServer
+    from transcribe.transcription import TranscriptionServer
     server = TranscriptionServer()
     server.run(
         "0.0.0.0",

transcribe/{translate.py → translator.py}

@@ -11,7 +11,8 @@ class QwenTranslator:
         # n_gpu_layers=-1, # Uncomment to use GPU acceleration
         # seed=1337, # Uncomment to set a specific seed
         # n_ctx=2048, # Uncomment to increase the context window
-        chat_format="chatml"
+        chat_format="chatml",
+        verbose=False
 )
         self.sys_prompt = system_prompt
 
@@ -27,10 +28,10 @@ class QwenTranslator:
         message = self.to_message(prompt, src_lang, dst_lang)
         start_time = time.monotonic()
         output = self.llm.create_chat_completion(messages=message, temperature=0.9)
-        logger.info(f"LLM translate cose: {time.monotonic() - start_time}s.")
-        return output['choices'][0]['content']
+        logger.info(f"LLM translate cose: {time.monotonic() - start_time:.2f}s.")
+        return output['choices'][0]['message']['content']
     
-    def __call__(self, prompt, max_tokens=256,*args, **kwargs):
+    def __call__(self, prompt, max_tokens=512,*args, **kwargs):
         return self.llm(
             prompt,
             *args,

transcribe/whisper_llm_serve.py

@@ -1,6 +1,5 @@
 
-from .server import ServeClientBase
-from pywhispercpp.model import Model
+
 import soundfile
 from concurrent.futures import ProcessPoolExecutor as Pool
 import numpy as np
@@ -12,7 +11,11 @@ import time
 import json
 import threading
 
-logger = getLogger("Pywhisper")
+from .server import ServeClientBase
+from .translator import QwenTranslator
+from pywhispercpp.model import Model
+
+logger = getLogger(__name__)
 
 
 class TripleTextBuffer:
@@ -102,13 +105,13 @@ class SegmentManager:
         
 
 class PywhisperInference:
-
-    model = None
+    whisper_model = None
+    llm_model = None
 
     @classmethod
     def initializer(cls, warmup=True):
         models_dir = config.MODEL_DIR.as_posix()
-        cls.model = Model(
+        cls.whisper_model = Model(
             model=config.WHISPER_MODEL, 
             models_dir=models_dir,
             print_realtime=False,
@@ -117,12 +120,15 @@ class PywhisperInference:
         )
         if warmup:
             cls.warmup()
+        
+        # init llamacpp
+        cls.llm_model = QwenTranslator(config.LLM_MODEL_PATH, config.LLM_SYS_PROMPT)
     
     @classmethod
     def warmup(cls, warmup_steps=1):
         mel, _, = soundfile.read("assets/jfk.flac")
         for _ in range(warmup_steps):
-            cls.model.transcribe(mel, print_progress=False)
+            cls.whisper_model.transcribe(mel, print_progress=False)
 
     @staticmethod
     def config_language(language):
@@ -136,7 +142,7 @@ class PywhisperInference:
     def inference(cls, audio_buffer, language):
         max_len, prompt = cls.config_language(language)
         audio_buffer = np.frombuffer(audio_buffer, dtype=np.float32)
-        return cls.model.transcribe(
+        return cls.whisper_model.transcribe(
             audio_buffer,
             initial_prompt=prompt,
             language=language,  
@@ -144,13 +150,18 @@ class PywhisperInference:
             max_len=max_len
         )
     
+    @classmethod
+    def translate(cls, context: str, src_lang, dst_lang):
+        return cls.llm_model.translate(context, src_lang, dst_lang)
+
 
 class PyWhiperCppServe(ServeClientBase):
    
     
-    def __init__(self, websocket, language=None, client_uid=None,):
+    def __init__(self, websocket, language=None, dst_lang=None, client_uid=None,):
         super().__init__(client_uid, websocket)
         self.language = language
+        self.dst_lang = dst_lang # 目标翻译语言
         # 设置观察字符串 对比上下次的文字来判断字符串的输出是否固定
         self._text_buffer = TripleTextBuffer()
         # 存储转录数据
@@ -162,8 +173,9 @@ class PyWhiperCppServe(ServeClientBase):
         self._pool = Pool(
             max_workers=1, initializer=PywhisperInference.initializer)
 
-        logger.info('Create a thread to process audio.')
+        logger.info('Create a process to process audio.')
         self.trans_thread = threading.Thread(target=self.speech_to_text)
+        self.trans_thread.daemon = True
         self.trans_thread.start()
 
         self.websocket.send(json.dumps({
@@ -191,9 +203,18 @@ class PyWhiperCppServe(ServeClientBase):
         Args:
             audio_buffer (np.array): The audio chunk to transcribe.
         """
-        fut = self._pool.submit(
+        transcribe_fut = self._pool.submit(
             PywhisperInference.inference, audio_buffer.tobytes(), self.language)
-        return fut.result()
+        segments = transcribe_fut.result()
+        
+        return segments
+    
+    def translate_text(self, text):
+        """
+        translate the text to dst lang"""
+        translate_fut = self._pool.submit(
+            PywhisperInference.translate, text, self.language, self.dst_lang)
+        return  translate_fut.result()
 
 
      
@@ -264,7 +285,8 @@ class PyWhiperCppServe(ServeClientBase):
             except Exception as e:
                 logger.error(f"[ERROR]: {e}")
             else:
-                self.handle_transcription_output(segments, audio_buffer)
+                for item in self.handle_transcription_output(segments, audio_buffer):
+                    print(item)
     
            
     
@@ -283,10 +305,8 @@ class PyWhiperCppServe(ServeClientBase):
         if is_end_sentence and last_cut_index:
             message = self._segment_manager.segment
             seg_id = self._segment_manager.get_seg_id() - 1
-            # elapsed_time = time.time() - start_time
-            # formatted_time = f"{int(elapsed_time // 60):02}:{int(elapsed_time % 60):02}:{(elapsed_time % 1) * 1000:03.0f}"
-            print(seg_id, message,)
-            print(seg_id + 1, self._segment_manager.string,)
+            yield (seg_id, message, self.translate_text(message))
+            yield (seg_id + 1, self._segment_manager.string, self.translate_text(self._segment_manager.string))
         
         else:
             seg_id = self._segment_manager.get_seg_id()
@@ -294,7 +314,7 @@ class PyWhiperCppServe(ServeClientBase):
             # print(self._segment_manager.__dict__)
             # elapsed_time = time.time() - start_time
             # formatted_time = f"{int(elapsed_time // 60):02}:{int(elapsed_time % 60):02}:{(elapsed_time % 1) * 1000:03.0f}"
-            print(seg_id, message)
+            yield (seg_id, message,self.translate_text(message))
     
     def send_to_client(self, data_dict):
         content = {

uv.lock

@@ -214,6 +214,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/4e/8c/f3147f5c4b73e7550fe5f9352eaa956ae838d5c51eb58e7a25b9f3e2643b/decorator-5.2.1-py3-none-any.whl", hash = "sha256:d316bb415a2d9e2d2b3abcc4084c6502fc09240e292cd76a76afc106a1c8e04a", size = 9190 },
 ]
 
+[[package]]
+name = "diskcache"
+version = "5.6.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/3f/21/1c1ffc1a039ddcc459db43cc108658f32c57d271d7289a2794e401d0fdb6/diskcache-5.6.3.tar.gz", hash = "sha256:2c3a3fa2743d8535d832ec61c2054a1641f41775aa7c556758a109941e33e4fc", size = 67916 }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/3f/27/4570e78fc0bf5ea0ca45eb1de3818a23787af9b390c0b0a0033a1b8236f9/diskcache-5.6.3-py3-none-any.whl", hash = "sha256:5e31b2d5fbad117cc363ebaf6b689474db18a1f6438bc82358b024abd4c2ca19", size = 45550 },
+]
+
 [[package]]
 name = "filelock"
 version = "3.18.0"
@@ -321,6 +330,18 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/b5/ba/c63c5786dfee4c3417094c4b00966e61e4a63efecee22cb7b4c0387dda83/librosa-0.11.0-py3-none-any.whl", hash = "sha256:0b6415c4fd68bff4c29288abe67c6d80b587e0e1e2cfb0aad23e4559504a7fa1", size = 260749 },
 ]
 
+[[package]]
+name = "llama-cpp-python"
+version = "0.3.8"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "diskcache" },
+    { name = "jinja2" },
+    { name = "numpy" },
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/95/4e/da912ff2bf9bf855c86e8b1ae9fe1eaedf47d75a66728896b533901c4610/llama_cpp_python-0.3.8.tar.gz", hash = "sha256:31c91323b555c025a76a30923cead9f5695da103dd68c15cdbb4509b17f0ed77", size = 67301056 }
+
 [[package]]
 name = "llvmlite"
 version = "0.44.0"
@@ -751,6 +772,37 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/5a/dc/491b7661614ab97483abf2056be1deee4dc2490ecbf7bff9ab5cdbac86e1/pyreadline3-3.5.4-py3-none-any.whl", hash = "sha256:eaf8e6cc3c49bcccf145fc6067ba8643d1df34d604a1ec0eccbf7a18e6d3fae6", size = 83178 },
 ]
 
+[[package]]
+name = "pywhispercpp"
+version = "1.3.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "numpy" },
+    { name = "platformdirs" },
+    { name = "requests" },
+    { name = "tqdm" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/85/bf/213ab706513a2ad6f0266b44c841f5123ae3592e769488be1e62268f0451/pywhispercpp-1.3.0.tar.gz", hash = "sha256:d5d4ddfdbe7472736343413355f47a74783bc40891c5826b607eac4f62b08279", size = 1580900 }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/79/4d/4765b2477e028d27f0f497f9e7e27ecfd207cf1d133c12626a7901ceca8a/pywhispercpp-1.3.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:f66cc5ff624744cf0b7da1c53d67d2e3416f92a02c7ae945697d47d13eed5845", size = 1562728 },
+    { url = "https://files.pythonhosted.org/packages/e5/0d/a17a403feeea0e0d064fb230d56119718ca813d6b382804848c387387022/pywhispercpp-1.3.0-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b2fb92868fb7219478f6bdff50d5045583466985e3e9017af040735eb650772d", size = 1893935 },
+    { url = "https://files.pythonhosted.org/packages/3f/09/db14ce3c4555a642ba62e4da226e578ea86d1c32ca70b93e32dae3309cf2/pywhispercpp-1.3.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:66270fb0dbb00745419001eef7c46c49e2104b80ebd21d8e65d58f0bdef7c2a7", size = 1788858 },
+    { url = "https://files.pythonhosted.org/packages/a0/28/fd4de678cfc50ad8dedba1ab5f9a3e3563759ca3dce273eb2b09b95e4729/pywhispercpp-1.3.0-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:f61f2e36f06bb98cd139aee073d6af5223fe551bab80ce3cbdc643eb7edb5e1b", size = 3026209 },
+    { url = "https://files.pythonhosted.org/packages/94/a3/81c3616eb033344e4af356896aded58c1f36fdca253294dd501db7630a09/pywhispercpp-1.3.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:bf0478a443ea7a56f35fbeba1c5f1988b13088aafc79d533a67020ee7ce146f3", size = 2792276 },
+    { url = "https://files.pythonhosted.org/packages/49/83/0bcb864c5dff802afd4c1887149bf029ea6715e20438feb0eb0727e0e70d/pywhispercpp-1.3.0-cp311-cp311-win32.whl", hash = "sha256:0c906e7537d9e81813d86bcc98824085058174b238900bad655c8d7bd426f9a4", size = 731243 },
+    { url = "https://files.pythonhosted.org/packages/a7/15/5421c0814e611aae9f1541c36d4bfba157ae30580424091519aaeffe3f1d/pywhispercpp-1.3.0-cp311-cp311-win_amd64.whl", hash = "sha256:3af25987a0d3caa88a6c795b2e0051b74515a81c33bdbd49c6e8ca65b7fdc6db", size = 863874 },
+    { url = "https://files.pythonhosted.org/packages/4b/e6/c4008fed98e0ab047bd458aaeb675226d2bd630c11939e641c3385657552/pywhispercpp-1.3.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:dd0903d9685447700ed4cad48f298cb73dd073bcd0abb8bcf974646006f4e222", size = 1563954 },
+    { url = "https://files.pythonhosted.org/packages/d9/31/5de448da80b8e58bda47e2ca503b97484afafe93d93793f5f7b6140f9fba/pywhispercpp-1.3.0-cp312-cp312-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:e84926618d819aaeb128170351b7489dc01cabb0e58c44c6b11f8449f0850d7a", size = 1893782 },
+    { url = "https://files.pythonhosted.org/packages/5c/82/cb38ffa05a0fcbff7547a42735b05f36d92af5352ae7273764e05ca85363/pywhispercpp-1.3.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2ba675b98ffd1901d3696613655700d3d407bd952e110a1bfbbb618997818d8b", size = 1787927 },
+    { url = "https://files.pythonhosted.org/packages/90/9d/3aa981b0551d44339ffc613c4466df4d8e3c59609da444a490eac032e0a2/pywhispercpp-1.3.0-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:2b8407a9cae5555bd276fac578155de6273f3bcc0ee8960c1fd0cfb7f4be1be9", size = 3027385 },
+    { url = "https://files.pythonhosted.org/packages/b0/3f/85924b54772e5ce5222bbf9aa98247800c7be2262daf17d07e2fae6e1ced/pywhispercpp-1.3.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:fd31fa0992a39e55d4a793dad1f6d2a1a97261998d52726f4568cea0af78ff93", size = 2791659 },
+    { url = "https://files.pythonhosted.org/packages/71/63/0ed9a4a776d81018536f2e5556fe364599a661aeded42627f93e1d4282aa/pywhispercpp-1.3.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:e8d1b339409668382f68b810c2ccac010e71d580c6afe6d67e86336daeb939ed", size = 1563943 },
+    { url = "https://files.pythonhosted.org/packages/e4/3b/f9fda6bd691900e76702945405919174838497096961f56aa8583b6373ed/pywhispercpp-1.3.0-cp313-cp313-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:0ac9fea9e3aeb5219ce614f644c4f1b38b9561eede1e4380ab39de96afba8864", size = 1893777 },
+    { url = "https://files.pythonhosted.org/packages/42/fc/26b0811b3964f4b14691ff28d84ccfd5fb7b800eab7ca898f538271c9c61/pywhispercpp-1.3.0-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fc6801332deb2d6efc67e5dd7c9c9b1ce06f8235270b3ebd5546a79040a67314", size = 1787957 },
+    { url = "https://files.pythonhosted.org/packages/3a/f2/1673b43721246a9da0a9820abb6aa442c4de0129d2039fd107f0657878f4/pywhispercpp-1.3.0-cp313-cp313-musllinux_1_2_i686.whl", hash = "sha256:5bc5742787037f421f748ad463d85bce565287e77dfddaa5536d37d59ac44267", size = 3027419 },
+    { url = "https://files.pythonhosted.org/packages/d2/33/292a727bd3aa826c5981df32ce228f3e89edf4116ae4cc9d76ddedb9fb07/pywhispercpp-1.3.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:a6f7578fb9a8de4c8663323060c0715713168ae174ab022dfe9ad13b21223c3c", size = 2791670 },
+]
+
 [[package]]
 name = "requests"
 version = "2.32.3"
@@ -1011,9 +1063,11 @@ source = { virtual = "." }
 dependencies = [
     { name = "av" },
     { name = "librosa" },
+    { name = "llama-cpp-python" },
     { name = "numpy" },
     { name = "onnxruntime" },
     { name = "pyaudio" },
+    { name = "pywhispercpp" },
     { name = "setuptools" },
     { name = "soundfile" },
     { name = "torch" },
@@ -1026,9 +1080,11 @@ dependencies = [
 requires-dist = [
     { name = "av", specifier = ">=14.2.0" },
     { name = "librosa", specifier = ">=0.11.0" },
+    { name = "llama-cpp-python", specifier = ">=0.3.8" },
     { name = "numpy", specifier = ">=2.1.3" },
     { name = "onnxruntime", specifier = ">=1.21.0" },
     { name = "pyaudio", specifier = ">=0.2.14" },
+    { name = "pywhispercpp", specifier = ">=1.3.0" },
     { name = "setuptools", specifier = ">=78.1.0" },
     { name = "soundfile", specifier = ">=0.13.1" },
     { name = "torch", specifier = ">=2.6.0" },