记录一下 coqui-ai/TTS 启动 http 服务的方式

参考官方教程：https://docs.coqui.ai/en/latest/docker_images.html#start-a-server

持久化运行容器：

1. docker run -d --restart always -p 5002:5002 --entrypoint python3 ghcr.io/coqui-ai/tts-cpu TTS/server/server.py --model_name tts_models/en/ljspeech/vits

复制代码

运行后直接打开网页测试：http://your_ip:5002/
音频 wav 音频文件链接为：http://your_ip:5002/api/tts?speaker_id=p364&text=nice%20to%20meet%20you

镜像很大，拉取时你要忍一下：



如果你想运行多个 model，需要跑多个容器，然后用不同端口映射去提供服务

有些模型不支持服务器模式，例如 tts_models/en/multi-dataset/tortoise-v2，所以你运行容器，发现无法通过端口访问，那估计是模型不支持

如果你想在 ARM 电脑上测试，需要用下面的方式去仿真 AMD：

1. docker pull ghcr.io/coqui-ai/tts-cpu --platform linux/amd64
   
2. docker run --rm -it -p 5002:5002 --platform linux/amd64 ghcr.io/coqui-ai/tts-cpu

复制代码

以下是我打印出的模型列表，每个版本的镜像可能模型不一样

不推荐用 /fast_pitch 结尾的模型，这种可能是因为追求速度，声音马赛克感很重

对比了多个模型，这个效果最好：tts_models/en/ljspeech/vits

1. python3 TTS/server/server.py --list_models
   
2. 
   
3. 
   
4. Name format: type/language/dataset/model
   
5. 1: tts_models/multilingual/multi-dataset/xtts_v2
   
6. 2: tts_models/multilingual/multi-dataset/xtts_v1.1
   
7. 3: tts_models/multilingual/multi-dataset/your_tts
   
8. 4: tts_models/multilingual/multi-dataset/bark
   
9. 5: tts_models/bg/cv/vits
   
10. 6: tts_models/cs/cv/vits
    
11. 7: tts_models/da/cv/vits
    
12. 8: tts_models/et/cv/vits
    
13. 9: tts_models/ga/cv/vits
    
14. 10: tts_models/en/ek1/tacotron2
    
15. 11: tts_models/en/ljspeech/tacotron2-DDC
    
16. 12: tts_models/en/ljspeech/tacotron2-DDC_ph
    
17. 13: tts_models/en/ljspeech/glow-tts
    
18. 14: tts_models/en/ljspeech/speedy-speech
    
19. 15: tts_models/en/ljspeech/tacotron2-DCA
    
20. 16: tts_models/en/ljspeech/vits
    
21. 17: tts_models/en/ljspeech/vits--neon
    
22. 18: tts_models/en/ljspeech/fast_pitch
    
23. 19: tts_models/en/ljspeech/overflow
    
24. 20: tts_models/en/ljspeech/neural_hmm
    
25. 21: tts_models/en/vctk/vits
    
26. 22: tts_models/en/vctk/fast_pitch
    
27. 23: tts_models/en/sam/tacotron-DDC
    
28. 24: tts_models/en/blizzard2013/capacitron-t2-c50
    
29. 25: tts_models/en/blizzard2013/capacitron-t2-c150_v2
    
30. 26: tts_models/en/multi-dataset/tortoise-v2
    
31. 27: tts_models/en/jenny/jenny
    
32. 28: tts_models/es/mai/tacotron2-DDC
    
33. 29: tts_models/es/css10/vits
    
34. 30: tts_models/fr/mai/tacotron2-DDC
    
35. 31: tts_models/fr/css10/vits
    
36. 32: tts_models/uk/mai/glow-tts
    
37. 33: tts_models/uk/mai/vits
    
38. 34: tts_models/zh-CN/baker/tacotron2-DDC-GST
    
39. 35: tts_models/nl/mai/tacotron2-DDC
    
40. 36: tts_models/nl/css10/vits
    
41. 37: tts_models/de/thorsten/tacotron2-DCA
    
42. 38: tts_models/de/thorsten/vits
    
43. 39: tts_models/de/thorsten/tacotron2-DDC
    
44. 40: tts_models/de/css10/vits-neon
    
45. 41: tts_models/ja/kokoro/tacotron2-DDC
    
46. 42: tts_models/tr/common-voice/glow-tts
    
47. 43: tts_models/it/mai_female/glow-tts
    
48. 44: tts_models/it/mai_female/vits
    
49. 45: tts_models/it/mai_male/glow-tts
    
50. 46: tts_models/it/mai_male/vits
    
51. 47: tts_models/ewe/openbible/vits
    
52. 48: tts_models/hau/openbible/vits
    
53. 49: tts_models/lin/openbible/vits
    
54. 50: tts_models/tw_akuapem/openbible/vits
    
55. 51: tts_models/tw_asante/openbible/vits
    
56. 52: tts_models/yor/openbible/vits
    
57. 53: tts_models/hu/css10/vits
    
58. 54: tts_models/el/cv/vits
    
59. 55: tts_models/fi/css10/vits
    
60. 56: tts_models/hr/cv/vits
    
61. 57: tts_models/lt/cv/vits
    
62. 58: tts_models/lv/cv/vits
    
63. 59: tts_models/mt/cv/vits
    
64. 60: tts_models/pl/mai_female/vits
    
65. 61: tts_models/pt/cv/vits
    
66. 62: tts_models/ro/cv/vits
    
67. 63: tts_models/sk/cv/vits
    
68. 64: tts_models/sl/cv/vits
    
69. 65: tts_models/sv/cv/vits
    
70. 66: tts_models/ca/custom/vits
    
71. 67: tts_models/fa/custom/glow-tts
    
72. 68: tts_models/bn/custom/vits-male
    
73. 69: tts_models/bn/custom/vits-female
    
74. 70: tts_models/be/common-voice/glow-tts
    
75. 
    
76. Name format: type/language/dataset/model
    
77. 1: vocoder_models/universal/libri-tts/wavegrad
    
78. 2: vocoder_models/universal/libri-tts/fullband-melgan
    
79. 3: vocoder_models/en/ek1/wavegrad
    
80. 4: vocoder_models/en/ljspeech/multiband-melgan
    
81. 5: vocoder_models/en/ljspeech/hifigan_v2
    
82. 6: vocoder_models/en/ljspeech/univnet
    
83. 7: vocoder_models/en/blizzard2013/hifigan_v2
    
84. 8: vocoder_models/en/vctk/hifigan_v2
    
85. 9: vocoder_models/en/sam/hifigan_v2
    
86. 10: vocoder_models/nl/mai/parallel-wavegan
    
87. 11: vocoder_models/de/thorsten/wavegrad
    
88. 12: vocoder_models/de/thorsten/fullband-melgan
    
89. 13: vocoder_models/de/thorsten/hifigan_v1
    
90. 14: vocoder_models/ja/kokoro/hifigan_v1
    
91. 15: vocoder_models/uk/mai/multiband-melgan
    
92. 16: vocoder_models/tr/common-voice/hifigan
    
93. 17: vocoder_models/be/common-voice/hifigan
    
94. 
    
95. Name format: type/language/dataset/model
    
96. 1: voice_conversion_models/multilingual/vctk/freevc24

复制代码