modify readme for using efficient speech

README.md

@@ -14,14 +14,17 @@
 - 实时和离线功能
 - 使用[ChatGLM3](https://github.com/THUDM/ChatGLM3) 6B 4比特量化模型进行聊天交互
 - 使用[whisper.cpp](https://github.com/ggerganov/whisper.cpp)加速自动语音识别（ASR）
-- 使用[TTS](https://github.com/coqui-ai/TTS)进行文本转语音转换
+- 使用[EfficientSpeech](https://github.com/roatienza/efficientspeech)进行文本转语音转换
 
 ## 使用方法
 - 根据[Chatglm.cpp](chatglm.cpp.md)安装Chatglm.cpp
 - 下载ChatGLM3 6B-4bit模型 [model](https://huggingface.co/Xorbits/chatglm3-6B-GGML)
 - 安装 [whisper.cpp](https://github.com/ggerganov/whisper.cpp) 尽量选择BLAS编译，加速推理速度
+- 安装 [EfficientSpeech](./examples/efficientspeech/README.md)
 - 安装相关依赖
 `pip install -r requirements.txt`
+- 创建本地文字转语音服务（TTS）
+`cd examples/efficientspeech/ && sh es_tts_service.sh`
 - 修改模型存储路径并运行脚本
     ```
     vim examples/demo.sh 

README_en.md

@@ -15,15 +15,18 @@ Tested on M2 16G MacBook Air / (4070Ti) + 13600KF Ubuntu / (4070Ti) + 13600KF Wi
 - Real-time and offline functionality
 - Utilizes [ChatGLM3](https://github.com/THUDM/ChatGLM3) 6B 4-bit quantized model for chat interactions
 - Accelerated automatic speech recognition (ASR) with [whisper.cpp](https://github.com/ggerganov/whisper.cpp)
-- Text-to-speech (TTS) conversion using [TTS](https://github.com/coqui-ai/TTS)
+- Text-to-speech (TTS) conversion using [EfficientSpeech](https://github.com/roatienza/efficientspeech)
 
 ## Usage
 1. Follow the README of [Chatglm.cpp](chatglm.cpp.md) to install chatglm.cpp
 2. Download ChatGLM3 6B-4bit [model](https://huggingface.co/Xorbits/chatglm3-6B-GGML)
 3. Install [whisper.cpp](https://github.com/ggerganov/whisper.cpp) and compile with BLAS / CUBLAS can speed up the inference process
-4. Install requirements
+4. Install [EfficientSpeech](./examples/efficientspeech/README.md) for real-time TTS
+5. Install requirements
 `pip install -r requirements.txt`
-5. Modify the model path of the script and run it:
+6. Create local TTS service
+`cd examples/efficientspeech/ && sh es_tts_service.sh`
+7. Modify the model path of the script and run it:
     ```
     vim examples/demo.sh 
     cd examples && sh demo.sh

examples/cli_demo.py

@@ -2,8 +2,7 @@
 import argparse
 from pathlib import Path
 from typing import List
-from TTS.api import TTS
-import playsound
+import requests
 import numpy as np
 from scipy.io import wavfile
 import warnings
@@ -23,9 +22,7 @@
                                               /_/   /_/       
 """.strip("\n")
 WELCOME_MESSAGE = "Welcome to ChatGLM.CPP-based oral English bot! Ask whatever you want. Say 'clear' to clear context. Say 'stop' to exit."
-
-import requests
-url = 'http://localhost:5000/convert_text_to_speech'
+tts_url = 'http://localhost:5000/convert_text_to_speech'
 
 def main() -> None:
     parser = argparse.ArgumentParser()
@@ -47,7 +44,6 @@ def main() -> None:
     parser.add_argument("--runtime_dir", default=None, type=str, help="path to save chat audio and text file")
     parser.add_argument("--asr_main", default=None, type=str, help="path to run asr main program")
     parser.add_argument("--asr_model", default=None, type=str, help="path to load asr model")
-    parser.add_argument("--tts_model", default=None, type=str, help="path to load tts model")
     parser.add_argument("--input_device", default=0, type=int, help="the index of the input device")
     args = parser.parse_args()
 
@@ -58,7 +54,6 @@ def main() -> None:
     if args.sp:
         system = args.sp.read_text()
     os.makedirs(args.runtime_dir, exist_ok=True)
-    # tts = TTS(model_name=args.tts_model, progress_bar=False)
     sampling_rate = 22050            
     sd.default.samplerate = sampling_rate
     sd.default.latency = 'low'
@@ -149,7 +144,7 @@ def main() -> None:
         ai_audio_path = f'{args.runtime_dir}/chat_{talk_round}_ai.wav'  
         response_txt = remove_non_english_chars(msg_out.content.replace('\n', ' ').strip())
         if response_txt:
-            response = requests.post(url, json={'text': response_txt})
+            response = requests.post(tts_url, json={'text': response_txt})
             if response.status_code == 200:
                 wav = response.json()['audio']
                 wav = np.array([x[0] for x in wav], dtype=np.float32).reshape((-1, 1))
@@ -158,8 +153,6 @@ def main() -> None:
                 sd.wait()
             else:
                 print('Error:', response.json())
-            # tts.tts_to_file(response_txt, file_path=ai_audio_path)
-            # playsound.playsound(ai_audio_path)
 
     print("Bye")
 

examples/demo.sh

@@ -2,13 +2,10 @@
 CHAT_MODEL=/Users/xuliu/Documents/vscode/llm/model_zoo/chatglm3-ggml-q4_0.bin
 ASR_MAIN=/Users/xuliu/Documents/vscode/speech/whisper.cpp/main
 ASR_MODEL=/Users/xuliu/Documents/vscode/speech/whisper.cpp/models/ggml-base.en.bin
-TTS_MODEL=tts_models/en/ljspeech/vits--neon
 python3 cli_demo.py \
   --model ${CHAT_MODEL} \
   --interactive \
   --runtime_dir './runtime' \
   --asr_main ${ASR_MAIN} \
   --asr_model ${ASR_MODEL} \
-  --tts_model ${TTS_MODEL} \
   --input_device 1
-

requirements.txt

@@ -2,4 +2,3 @@ pynput
 scipy
 sounddevice
 subprocess
-TTS