Text to Speech API

Documentation for TTS

Project Documentation: https://open-speech-ekstep.github.io/tts_model_api/

Api Documentation: Swagger API document is hosted on /docs endpoint.

Sample Request:

{
    "input": [
        {
            "source": "भारत मेरा देश है|"
        }
    ],
    "config": {
        "gender": "female",
        "language": {
            "sourceLanguage": "hi"
        }
    }
}

Sample Response:

{
    "audio": [
        {
            "audioContent": "<Audio Bytes>"
        }
    ],
    "config": {
        "language": {
            "sourceLanguage": "hi"
        },
        "audioFormat": "wav",
        "encoding": "base64",
        "samplingRate": 22050
    }
}

Supported environment variables

Variable Name	Default Value	Description
server_port	5000	Port for unicorn server
log_level	DEBUG	Log level for application logs
gpu	True	True: Load the models on GPU False: Use CPU
tts_max_text_limit	450	Maximum length of text that will be sent to model. Text above this will be broken in chunks before hitting model.
models_base_path		Location for language model folders.
model_config_file_path	model_dict.json

Running with Docker image

You can use a pre-built image of this repo to run in docker. Models are open sourced in our https://github.com/Open-Speech-EkStep/vakyansh-models.

Pre-built docker images are hosted on gcr.io/ekstepspeechrecognition/speech_recognition_model_api. We do not follow the latest tag, so you have to use a specific tag.

docker pull gcr.io/ekstepspeechrecognition/text_to_speech_open_api:2.1.15

Running with docker

docker run -itd -p 5000:5000 --gpus all -v <your location for deployed_models>:/opt/text_to_speech_open_api/deployed_models/ -v <your location for translit_models>:/opt/text_to_speech_open_api/vakyansh-tts/src/glow_tts/tts_infer/translit_models/ gcr.io/ekstepspeechrecognition/text_to_speech_open_api:2.1.15

Directory structure of /opt/text_to_speech_open_api/deployed_models/

.
|-- gujarati
|   |-- female
|   |   |-- glow_tts
|   |   |   |-- G_190.pth
|   |   |   `-- config.json
|   |   `-- hifi_tts
|   |       |-- config.json
|   |       `-- g_00100000
|   `-- male
|       |-- glow_tts
|       |   |-- G_170.pth
|       |   `-- config.json
|       `-- hifi_tts
|           |-- config.json
|           `-- g_00100000
|-- hindi
|   |-- female
|   |   |-- glow_tts
|   |   |   |-- G_100.pth
|   |   |   |-- G_250.pth
|   |   |   `-- config.json
|   |   `-- hifi_tts
|   |       |-- config.json
|   |       `-- g_00100000
|   `-- male
|       |-- glow_tts
|       |   |-- G_100.pth
|       |   `-- config.json
|       `-- hifi_tts
|           |-- config.json
|           `-- g_00200000

Directory structure for /opt/text_to_speech_open_api/vakyansh-tts/src/glow_tts/tts_infer/translit_models/ translit models will be downloaded automatically.

.
|-- README.md
|-- bengali
|   |-- bn_101_model.pth
|   |-- bn_scripts.json
|   `-- bn_words_a4b.json
|-- default_lineup.json
|-- gujarati
|   |-- gu_101_model.pth
|   |-- gu_scripts.json
|   `-- gu_words_a4b.json
|-- hindi
|   |-- hi_111_model.pth
|   |-- hi_scripts.json
|   |-- hi_words_a4b.json

Sample model_dict.json

{
  "hi": {
    "male_glow": "hindi/male/glow_tts",
    "male_hifi": "hindi/male/hifi_tts",
    "female_glow": "hindi/female/glow_tts",
    "female_hifi": "hindi/female/hifi_tts"
  },
  "ta": {
    "male_glow": "tamil/male/glow_tts",
    "male_hifi": "tamil/male/hifi_tts",
    "female_glow": "tamil/female/glow_tts",
    "female_hifi": "tamil/female/hifi_tts"
  }
}

Building from source

We build this app in two steps to expedite the process of changes in the main source. We build a dependency image gcr.io/ekstepspeechrecognition/text_to_speech_open_api_dependency for which you can find dependency docker image file at dependencies/Dockerfile. Using dependency image, we build the main images which are published at gcr.io/ekstepspeechrecognition/text_to_speech_open_api. Docker file for this step is available here. You can use these steps to recreate the bundle. We recommend using some environment manager like conda.