一个用于 Deepgram 的语音转文本和文本转语音 API 的 Dart 和 Flutter 客户端,可在 Android、iOS、Web、macOS、Windows 和 Linux 上实现实时和批量转录。它支持流式音频输入、可自定义的转录设置,并可与语音助手集成。该包简化了在跨平台应用中访问高级语音识别和合成功能的过程。
适合场景
实时语音转录
语音助手集成
录音音频的转录
多语言语音处理
实时字幕
音频内容索引
主要特点
跨平台支持
流式音频转录
可自定义的转录参数
与 Deepgram 的 AI 模型集成
对多种语言的支持
文本转语音功能
适合对象
开发语音功能应用的开发者
需要准确语音识别的应用程序
跨平台音频处理
需要实时转录的应用程序
集成AI驱动的语音服务
使用建议
需要 Deepgram API 密钥
依赖网络连接
可能会产生基于使用的费用
有限的离线功能
平台特定权限需要
原始 README
以下为关联源码仓库的 README 原文快照,最新内容请访问源码仓库。
Deepgram Dart
A Deepgram client for Dart and Flutter, supporting all Speech-to-Text and Text-to-Speech features on every platform.
Feel free to create issues, contribute to this project or to ask for new features on GitHub !
Features
Speech-to-Text
Status
Methods
From File
✅
listen.file(), listen.path()
From URL
✅
listen.url()
From Byte
✅
listen.bytes()
From Audio Stream
✅
listen.live(), listen.liveListener()
Text-to-Speech
Status
Methods
From Text
✅
speak.text()
From Text Stream
✅
speak.live(), speak.liveSpeaker()
Agent Interaction
Status
Methods
Agent Interaction
🚧
agent.live()
PRs are welcome for all work-in-progress 🚧 features
Getting started
All you need is a Deepgram API key. You can get a free one by signing up on Deepgram
Usage
First create the client with your api key.
Deepgram deepgram = Deepgram('your_api_key');
Pre-recorded
Then you can call the methods you need under the propper listen or speak subclass:
// Speech to textDeepgramListenResult res = await deepgram.listen.file(File('audio.wav'), queryParams: {
'model': 'nova-2-general',
'detect_language': true,
'filler_words': false,
'punctuate': true,
// options here : https://developers.deepgram.com/reference/listen-file
});
// Text to speechDeepgramSpeakResult res = await deepgram.speak.text('Hello world', queryParams: {
'model': 'aura-asteria-en'// options here : https://developers.deepgram.com/reference/text-to-speech-api
});
Streaming
Speech to text
let's say from a microphone :
// https://pub.dev/packages/record (other packages would work too)Stream<List<int>> micStream = awaitAudioRecorder().startStream(RecordConfig(
encoder: AudioEncoder.pcm16bits,
sampleRate: 16000,
numChannels: 1,
));
final sttStreamParams = {
'detect_language': false, // not supported by streaming API'language': 'en',
// must specify encoding and sample_rate according to the audio stream'encoding': 'linear16',
'sample_rate': 16000,
};
then you got 2 options depending if you want to have more control over the stream or not :
// 1. you want the stream to manage itself automaticallyStream<DeepgramListenResult> stream = deepgram.listen.live(micStream, queryParams: sttStreamParams);
// 2. you want to manage the stream manuallyDeepgramLiveListener listener = deepgram.liveListener(micStream, queryParams: sttStreamParams);
listener.stream.listen((res) {
print(res.transcript);
});
// connect to the servers and start sending data
listener.start();
// you can pause and resume the transcription (stop sending audio data to the server)
listener.pause();
// ...
listener.resume();
// then close the stream when you're done, you can call start() again if you want to restart a transcription
listener.close();
final textStream = ...
// 1. you want the stream to manage itself automaticallyStream<DeepgramSpeakResult> stream = deepgram.speak.live(textStream, queryParams: ttsStreamParams);
// 2. you want to manage the stream manuallyDeepgramLiveSpeaker speaker = deepgram.liveListener(textStream, queryParams: ttsStreamParams);
speaker.stream.listen((res) {
print(res);
// if you want to use the audio, simplest way is to use Deepgram.toWav(res.data) !
});
// start sending data to the servers
speaker.start();
// https://developers.deepgram.com/docs/tts-ws-flush
speaker.flush();
//https://developers.deepgram.com/docs/tts-ws-clear
speaker.clear();
// then close the stream when you're done, you can call start() again if you want to restart a transcription
speaker.close();
Tested on Android and iOS, but should work on other platforms too.
Debugging common errors
make sure your API key is valid and has enough credits
deepgram.isApiKeyValid()
"Websocket was not promoted ..." : you are probably using wrong parameters, for example trying to use a whisper model with live streaming (not supported by deepgram)
empty transcript/only metadata : if streaming check that you specified encoding and sample_rate properly and that it matches the audio stream
double check the parameters you are using, some are not supported for streaming or for some models
Additional information
I created this package for my own needs since there are no dart sdk for deepgram. Happy to share !
Don't hesitate to ask for new features or to contribute on GitHub !
Support
If you'd like to support this project, consider contributing here. Thank you! :)