在Rust中使用portaudio从深度语音中获取空白结果

Question

我正试图将portaudio与深度语音结合使用(两者都使用Rust绑定(来创建语音识别程序。当我记录缓冲区时，我可以看到数据，但当尝试使用intermediate_decode时，我总是得到空白结果。我假设我要么配置音频不正确，要么设置模型不正确。我花了很多时间才走到这一步(对处理音频还很陌生(，如果有任何帮助，我将不胜感激！

这是完整的源代码：

use deepspeech::Model;
use portaudio as pa;
use std::path::Path;
fn start_recognition(mut model: Model) {
let pa = pa::PortAudio::new().expect("Unable to init PortAudio");
let input_settings = pa.default_input_stream_settings(1, 16000.0, 1024).unwrap();
let process_audio = move |pa::InputStreamCallbackArgs { buffer, .. }| {
let mut stream = model
.create_stream()
.expect("Failed to create model stream");
stream.feed_audio(&buffer);
let text = stream.intermediate_decode();
match text {
Ok(t) => {
if t.chars().count() > 0 {
println!("Text: {}", t)
}
pa::Continue
}
Err(err) => {
eprintln!("Error: {:?}", err);
pa::Complete
}
}
};
let mut stream = pa
.open_non_blocking_stream(input_settings, process_audio)
.expect("Unable to create audio stream");
stream.start().expect("Unable to start audio stream");
while let true = stream.is_active().unwrap() {}
}
fn get_model() -> Model {
let dir_path = Path::new("src/models");
let mut graph_name: Box<Path> = dir_path.join("output_graph.pb").into_boxed_path();
for file in dir_path
.read_dir()
.expect("Specified model dir is not a dir")
{
if let Ok(f) = file {
let file_path = f.path();
if file_path.is_file() {
if let Some(ext) = file_path.extension() {
if ext == "pb" || ext == "pbmm" {
graph_name = file_path.into_boxed_path();
}
}
}
}
}
Model::load_from_files(&graph_name).unwrap()
}
fn main() {
let model = get_model();
start_recognition(model);
}

Answer 1

原来问题出在process_audio回调上。我需要将模型流的初始化转移到回调之外。

fn start_recognition(mut model: Model) {
let pa = pa::PortAudio::new().expect("Unable to init PortAudio");
let input_settings = pa.default_input_stream_settings(1, 16000.0, 1024).unwrap();
let mut stream = model
.create_stream()
.expect("Failed to create model stream");
let process_audio = move |pa::InputStreamCallbackArgs { buffer, .. }| {
stream.feed_audio(&buffer);
let text = stream.intermediate_decode();
match text {
Ok(t) => {
if t.chars().count() > 0 {
println!("Text: {}", t)
}
pa::Continue
}
Err(err) => {
eprintln!("Error: {:?}", err);
pa::Complete
}
}
};
let mut stream = pa
.open_non_blocking_stream(input_settings, process_audio)
.expect("Unable to create audio stream");
stream.start().expect("Unable to start audio stream");
while let true = stream.is_active().unwrap() {}
}