m

KotobaASRLLM/convert_ts_to_wav.py

@@ -0,0 +1,98 @@
+import os
+import glob
+import subprocess
+import re
+
+def sort_key(filename):
+    # Extract number from filename like "tbs-38532.ts"
+    match = re.search(r'(\d+)', filename)
+    if match:
+        return int(match.group(1))
+    return filename
+
+def main():
+    # Base directory containing the script
+    base_dir = os.path.dirname(os.path.abspath(__file__))
+    temp_video_dir = os.path.join(base_dir, "TempVideo")
+    
+    # Check if directory exists
+    if not os.path.exists(temp_video_dir):
+        print(f"Error: Directory {temp_video_dir} does not exist.")
+        return
+
+    # Find all .ts files
+    ts_files = glob.glob(os.path.join(temp_video_dir, "*.ts"))
+    if not ts_files:
+        print("No .ts files found in TempVideo directory.")
+        return
+
+    # Sort files to ensure correct order
+    ts_files.sort(key=lambda x: sort_key(os.path.basename(x)))
+    
+    wav_files = []
+    
+    print(f"Found {len(ts_files)} .ts files. Starting conversion...")
+
+    for ts_file in ts_files:
+        wav_file = os.path.splitext(ts_file)[0] + ".wav"
+        wav_files.append(wav_file)
+        
+        # Check if wav already exists to avoid re-encoding (optional, but good for retries)
+        # For now, we overwrite as per request implying a fresh run, or use -y in ffmpeg
+        
+        cmd = [
+            "ffmpeg",
+            "-y", # Overwrite output files
+            "-i", ts_file,
+            "-ar", "16000",
+            "-ac", "1",
+            "-c:a", "pcm_s16le",
+            wav_file
+        ]
+        
+        print(f"Converting {os.path.basename(ts_file)} to {os.path.basename(wav_file)}...")
+        try:
+            subprocess.run(cmd, check=True, stdout=subprocess.DEVNULL, stderr=subprocess.STDOUT)
+        except subprocess.CalledProcessError as e:
+            print(f"Error converting {ts_file}: {e}")
+            return
+        except FileNotFoundError:
+            print("Error: ffmpeg not found. Please ensure ffmpeg is installed and in your PATH.")
+            return
+
+    print("All conversions complete. Merging files...")
+    
+    # Create concat list file
+    concat_list_path = os.path.join(temp_video_dir, "concat_list.txt")
+    with open(concat_list_path, "w", encoding="utf-8") as f:
+        for wav_file in wav_files:
+            # Escape single quotes in path if necessary (though simple filenames usually don't need it)
+            # ffmpeg requires paths to be escaped
+            safe_path = wav_file.replace("'", "'\\''")
+            f.write(f"file '{safe_path}'\n")
+            
+    merged_output_path = os.path.join(temp_video_dir, "merged_output.wav")
+    
+    # Merge command
+    merge_cmd = [
+        "ffmpeg",
+        "-y",
+        "-f", "concat",
+        "-safe", "0",
+        "-i", concat_list_path,
+        "-c", "copy",
+        merged_output_path
+    ]
+    
+    try:
+        subprocess.run(merge_cmd, check=True, stdout=subprocess.DEVNULL, stderr=subprocess.STDOUT)
+        print(f"Successfully merged into: {merged_output_path}")
+    except subprocess.CalledProcessError as e:
+        print(f"Error merging files: {e}")
+    
+    # Clean up list file
+    if os.path.exists(concat_list_path):
+        os.remove(concat_list_path)
+
+if __name__ == "__main__":
+    main()

KotobaASRLLM/readme.txt

@@ -0,0 +1,43 @@
+
+
+see 深入理解神经网络：从逻辑回归到CNN.md  -> kotoba-whisper-v2.0-ggml
+
+see https://huggingface.co/kotoba-tech/kotoba-whisper-v2.0-ggml
+
+see https://github.com/ggerganov/whisper.cpp
+
+see https://github.com/kotoba-tech/tts_eval/blob/main/tts_eval/metric_asr.py  kotoba-whisper-v2.0 + 语音相似度
+
+see huggingface_echodict\CefSharpPlayer\CefSharpPlayer\readme.txt IPTV 输出每个流的信息
+
+see huggingface_echodict\IPTV2\readme.txt 边下载边推流 + 伪推流
+
+see huggingface_echodict\IPTV2\TempVideo 下载视频
+
+
+
+cd kotoba-whisper-v2.0-ggml/whisper.cpp \
+  && make -j
+
+
+./whisper-cli.exe -m ggml-kotoba-whisper-v2.0.bin -l ja -f 60s.wav --output-file transcription --output-json
+./whisper-cli.exe -m ggml-kotoba-whisper-v2.0.bin -l ja -f segment_0001.wav --output-file transcription --output-json
+	成功识别
+
+
+ffprobe -print_format json  -show_streams KotobaASRLLM/TempVideo/tbs-38532.ts
+   输出每个流的信息
+    Stream #0:1[0x101]: Audio: aac (LC) ([15][0][0][0] / 0x000F), 48000 Hz, stereo, fltp, 130 kb/s, start 62720.902889
+
+ffmpeg -i KotobaASRLLM/TempVideo/tbs-38532.ts -c:a pcm_s16le -ar 16000 -ac 1 -y KotobaASRLLM/TempVideo/tbs-38532.wav
+  转成 wav 
+  采样率 ： 16000 Hz (已调整为 ASR 常用标准)
+  声道 ： 单声道 (mono) (已调整为 ASR 常用标准)
+
+Note that it runs only with 16-bit WAV files, so make sure to convert your input before running the tool. For example, you can use ffmpeg like this:
+
+ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav
+
+
+python convert_ts_to_wav.py
+  调用 ffmpeg 把这目录下的所有 ts 视频转成指定格式的 wav 音频，最后再合并所有 wav 音频片段成为一个长音频