add export data for test

.gitignore

@@ -173,3 +173,4 @@ cython_debug/
 
 .idea/
 pywhispercpp/
+test_data.csv

api_model.py

@@ -15,6 +15,21 @@ class TransResult(BaseModel):
         populate_by_name = True 
 
 
+class DebugResult(BaseModel):
+    # trans_pattern: str
+    seg_id: int
+    transcrible_time: float
+    translate_time:float
+    context: str = Field(alias="transcribleContent")
+    from_: str = Field(alias="from")
+    to: str
+    tran_content: str = Field(alias="translateContent")
+    partial: bool = True
+
+    class Config:
+        populate_by_name = True 
+
+
 
 
 class Message(BaseModel):

config.py

@@ -9,6 +9,8 @@ logging.basicConfig(
     datefmt="%H:%M:%S"
 )
 
+TEST = True
+
 logging.getLogger("pywhispercpp").setLevel(logging.WARNING)
 
 

transcribe/utils.py

@@ -5,7 +5,7 @@ import logging
 import numpy as np
 from scipy.io.wavfile import write
 import config
-
+import csv
 import av
 def log_block(key: str, value, unit=''):
     if config.DEBUG:
@@ -94,3 +94,24 @@ def resample(file: str, sr: int = 16000):
 
 def save_to_wave(filename, data:np.ndarray, sample_rate=16000):
     write(filename, sample_rate, data)
+
+
+class TestDataWriter:
+    def __init__(self, file_path='test_data.csv'):
+        self.file_path = file_path
+        self.fieldnames = [
+            'seg_id', 'transcrible_time', 'translate_time',
+            'transcribleContent', 'from', 'to', 'translateContent', 'partial'
+        ]
+        self._ensure_file_has_header()
+
+    def _ensure_file_has_header(self):
+        if not os.path.exists(self.file_path) or os.path.getsize(self.file_path) == 0:
+            with open(self.file_path, mode='w', newline='') as file:
+                writer = csv.DictWriter(file, fieldnames=self.fieldnames)
+                writer.writeheader()
+
+    def write(self, result: 'DebugResult'):
+        with open(self.file_path, mode='a', newline='') as file:
+            writer = csv.DictWriter(file, fieldnames=self.fieldnames)
+            writer.writerow(result.model_dump(by_alias=True))

transcribe/whisper_llm_serve.py

@@ -7,13 +7,15 @@ from logging import getLogger
 from typing import List, Optional, Iterator, Tuple, Any
 import asyncio
 import numpy as np
+import config
 # import wordninja
-from api_model import TransResult, Message
+from api_model import TransResult, Message, DebugResult
 from .server import ServeClientBase
-from .utils import log_block, save_to_wave
+from .utils import log_block, save_to_wave, TestDataWriter
 from .translatepipes import TranslatePipes
 from .strategy import (
     TranscriptStabilityAnalyzer, TranscriptToken)
+import csv
 
 logger = getLogger("TranscriptionService")
 
@@ -37,6 +39,7 @@ class WhisperTranscriptionService(ServeClientBase):
         self.frames_np = None
         self.lock = threading.Lock()
         self._frame_queue = queue.Queue()
+        
 
         # 文本分隔符，根据语言设置
         self.text_separator = self._get_text_separator(language)
@@ -47,10 +50,25 @@ class WhisperTranscriptionService(ServeClientBase):
         # 启动处理线程
         self._translate_thread_stop = threading.Event()
         self._frame_processing_thread_stop = threading.Event()
+
         self.translate_thread = self._start_thread(self._transcription_processing_loop)
         self.frame_processing_thread = self._start_thread(self._frame_processing_loop)
 
+        # for test
+        self._transcrible_time_cost = 0.
+        self._translate_time_cost = 0.
+        if config.TEST:
+            self._test_task_event = threading.Event()
+            self._test_queue = queue.Queue()
+            self._test_thread = self._start_thread(self.test_data_loop)
+          
         # self._c = 0
+    
+    def test_data_loop(self):
+        writer = TestDataWriter()
+        while not self._test_task_event.is_set():
+            test_data = self._test_queue.get()
+            writer.write(test_data)  # Save test_data to CSV
 
 
     def _start_thread(self, target_function) -> threading.Thread:
@@ -114,7 +132,13 @@ class WhisperTranscriptionService(ServeClientBase):
         """从音频缓冲区中移除已处理的部分"""
         with self.lock:
             if self.frames_np is not None and offset > 0:
+                # self._c += 1
+                # before = self.frames_np.copy()
                 self.frames_np = self.frames_np[offset:]
+                # after = self.frames_np.copy()
+                # save_to_wave(f"./tests/{self._c}_before_cut.wav", before)
+                # save_to_wave(f"./tests/{self._c}_after_cut.wav", after)
+
 
     def _get_audio_for_processing(self) -> Optional[np.ndarray]:
         """准备用于处理的音频块"""
@@ -147,11 +171,12 @@ class WhisperTranscriptionService(ServeClientBase):
 
         result = self._translate_pipe.transcrible(audio_buffer.tobytes(), self.source_language)
         segments = result.segments
+        time_diff = (time.perf_counter() - start_time)
         logger.debug(f"📝 Transcrible Segments: {segments} ")
         logger.debug(f"📝 Transcrible: {self.text_separator.join(seg.text for seg in segments)} ")
         log_block("📝 Transcrible output", f"{self.text_separator.join(seg.text for seg in segments)}", "")
-        log_block("📝 Transcrible time", f"{(time.perf_counter() - start_time):.3f}", "s")
-
+        log_block("📝 Transcrible time", f"{time_diff:.3f}", "s")
+        self._transcrible_time_cost = round(time_diff, 3)
         return [
             TranscriptToken(text=s.text, t0=s.t0, t1=s.t1)
             for s in segments
@@ -167,10 +192,10 @@ class WhisperTranscriptionService(ServeClientBase):
 
         result = self._translate_pipe.translate(text, self.source_language, self.target_language)
         translated_text = result.translate_content
-
-        log_block("🐧 Translation time ", f"{(time.perf_counter() - start_time):.3f}", "s")
+        time_diff = (time.perf_counter() - start_time)
+        log_block("🐧 Translation time ", f"{time_diff:.3f}", "s")
         log_block("🐧 Translation out ", f"{translated_text}")
-
+        self._translate_time_cost = round(time_diff, 3)
         return translated_text
 
     def _translate_text_large(self, text: str) -> str:
@@ -183,10 +208,10 @@ class WhisperTranscriptionService(ServeClientBase):
 
         result = self._translate_pipe.translate_large(text, self.source_language, self.target_language)
         translated_text = result.translate_content
-
-        log_block("Translation large model time ", f"{(time.perf_counter() - start_time):.3f}", "s")
+        time_diff = (time.perf_counter() - start_time)
+        log_block("Translation large model time ", f"{time_diff:.3f}", "s")
         log_block("Translation large model output", f"{translated_text}")
-
+        self._translate_time_cost = round(time_diff, 3)
         return translated_text
 
 
@@ -253,6 +278,17 @@ class WhisperTranscriptionService(ServeClientBase):
             )
             current_time = time.perf_counter()
             time_diff = current_time - start_time
+            if config.TEST:
+                self._test_queue.put(DebugResult(
+                    seg_id=ana_result.seg_id,
+                    transcrible_time=self._transcrible_time_cost,
+                    translate_time=self._translate_time_cost,
+                    context=ana_result.context,
+                    from_=self.source_language,
+                    to=self.target_language,
+                    tran_content=translated_context,
+                    partial=ana_result.partial()
+                ))
             log_block("🚦 Traffic times diff", round(time_diff, 2), 's')