langchain-rust 5.2.0

use std::{error::Error, sync::Arc};

use async_openai::types::audio::CreateSpeechRequestArgs;
use async_openai::Client;
pub use async_openai::{
    config::{Config, OpenAIConfig},
    types::audio::{SpeechModel, SpeechResponseFormat, Voice},
};
use async_trait::async_trait;
use serde_json::Value;

use crate::tools::{SpeechStorage, Tool};

#[derive(Clone)]
pub struct Text2SpeechOpenAI<C: Config> {
    config: C,
    model: SpeechModel,
    voice: Voice,
    storage: Option<Arc<dyn SpeechStorage>>,
    response_format: SpeechResponseFormat,
    path: String,
}

impl<C: Config> Text2SpeechOpenAI<C> {
    pub fn new(config: C) -> Self {
        Self {
            config,
            model: SpeechModel::Tts1,
            voice: Voice::Alloy,
            storage: None,
            response_format: SpeechResponseFormat::Mp3,
            path: "./data/audio.mp3".to_string(),
        }
    }

    pub fn with_model(mut self, model: SpeechModel) -> Self {
        self.model = model;
        self
    }

    pub fn with_voice(mut self, voice: Voice) -> Self {
        self.voice = voice;
        self
    }

    pub fn with_storage<SS: SpeechStorage + 'static>(mut self, storage: SS) -> Self {
        self.storage = Some(Arc::new(storage));
        self
    }

    pub fn with_response_format(mut self, response_format: SpeechResponseFormat) -> Self {
        self.response_format = response_format;
        self
    }

    pub fn with_path<S: Into<String>>(mut self, path: S) -> Self {
        self.path = path.into();
        self
    }

    pub fn with_config(mut self, config: C) -> Self {
        self.config = config;
        self
    }
}

impl Default for Text2SpeechOpenAI<OpenAIConfig> {
    fn default() -> Self {
        Self::new(OpenAIConfig::default())
    }
}

#[async_trait]
impl<C: Config + Clone + Send + Sync> Tool for Text2SpeechOpenAI<C> {
    fn name(&self) -> String {
        "Text2SpeechOpenAI".to_string()
    }

    fn description(&self) -> String {
        r#"A wrapper around OpenAI Text2Speech. "
        "Useful for when you need to convert text to speech. "
        "It supports multiple languages, including English, German, Polish, "
        "Spanish, Italian, French, Portuguese""#
            .to_string()
    }

    async fn run(&self, input: Value) -> Result<String, Box<dyn Error>> {
        let input = input.as_str().ok_or("Invalid input")?;
        let client = Client::with_config(self.config.clone());
        let response_format: SpeechResponseFormat = self.response_format;

        let request = CreateSpeechRequestArgs::default()
            .input(input)
            .voice(self.voice.clone())
            .response_format(response_format)
            .model(self.model.clone())
            .build()?;

        let response = client.audio().speech().create(request).await?;

        if self.storage.is_some() {
            let storage = self.storage.as_ref().unwrap(); //safe to unwrap
            let data = response.bytes;
            return storage.save(&self.path, &data).await;
        } else {
            response.save(&self.path).await?;
        }

        Ok(self.path.clone())
    }
}

#[cfg(test)]
mod tests {
    use crate::tools::{Text2SpeechOpenAI, Tool};

    #[tokio::test]
    #[ignore]
    async fn openai_speech2text_tool() {
        let openai = Text2SpeechOpenAI::default();
        let s = openai.call("Hola como estas").await.unwrap();
        println!("{}", s);
    }

    #[tokio::test]
    async fn speech_uses_configured_client_and_writes_file() {
        use super::*;
        use serde_json::json;

        let mut server = mockito::Server::new_async().await;
        let mock = server
            .mock("POST", "/audio/speech")
            .match_header("authorization", "Bearer test-key")
            .match_body(mockito::Matcher::PartialJson(json!({
                "input": "hola",
                "model": "tts-1-hd",
                "voice": "nova",
                "response_format": "wav"
            })))
            .with_body([1u8, 2, 3])
            .create_async()
            .await;

        let nanos = std::time::SystemTime::now()
            .duration_since(std::time::UNIX_EPOCH)
            .unwrap()
            .as_nanos();
        let dir = std::env::temp_dir().join(format!("lcr-tts-{}-{nanos}", std::process::id()));
        let path = dir.join("out.wav");
        let tool = Text2SpeechOpenAI::new(
            OpenAIConfig::new()
                .with_api_base(server.url())
                .with_api_key("test-key"),
        )
        .with_model(SpeechModel::Tts1Hd)
        .with_voice(Voice::Nova)
        .with_response_format(SpeechResponseFormat::Wav)
        .with_path(path.to_string_lossy());

        let out = tool.run(json!("hola")).await.unwrap();
        mock.assert_async().await;
        assert_eq!(out, path.to_string_lossy());
        assert_eq!(std::fs::read(&path).unwrap(), vec![1, 2, 3]);
        let _ = std::fs::remove_dir_all(dir);
    }
}