index tts stash save

tests/messy_test.rs

@@ -1,22 +1,31 @@
 // use std::io::Cursor;
 
-use aha::utils::audio_utils::create_hann_window;
+use std::time::Instant;
+
+use aha::utils::{audio_utils::create_hann_window, tensor_utils::interpolate_nearest_1d};
 use anyhow::Result;
-use candle_core::DType;
+use candle_core::{DType, Tensor};
 // use symphonia::core::io::MediaSourceStream;
 
 #[test]
 fn messy_test() -> Result<()> {
     // RUST_BACKTRACE=1 cargo test -F cuda,ffmpeg messy_test -r -- --nocapture
     let device = &candle_core::Device::Cpu;
+    let t = Tensor::arange(0.0f32, 40.0, device)?.broadcast_as((1, 40, 40))?;
+    println!("t: {}", t);
+    let i_start = Instant::now();
+    let t_inter = interpolate_nearest_1d(&t, 20)?;
+    let i_duration = i_start.elapsed();
+    println!("Time elapsed in interpolate_nearest_1d is: {:?}", i_duration);
+    println!("t_inter: {}", t_inter);    
     // let url = "https://sis-sample-audio.obs.cn-north-1.myhuaweicloud.com/16k16bit.mp3";
     // let client = reqwest::blocking::Client::new();
     // let response = client.get(url).send()?;
     // let vec_u8 = response.bytes()?.to_vec();
     // let mut content = Cursor::new(vec_u8);
     // let mss = MediaSourceStream::new(Box::new(content), Default::default());
-    let window = create_hann_window(400, DType::F32, device)?;
-    println!("window: {}", window);
+    // let window = create_hann_window(400, DType::F32, device)?;
+    // println!("window: {}", window);
     // let audio_path = "file:///home/jhq/Videos/voice_01.wav";
     // let audio_path = "/home/jhq/Videos/zh.mp3";
     // let audio_path = "/home/jhq/Videos/zh.mp3";

tests/test_index_tts2.rs

@@ -0,0 +1,48 @@
+use std::time::Instant;
+use anyhow::Result;
+
+use aha::models::index_tts2::{generate::IndexTTS2Generate, utils::download_index_tts2_need_model};
+use aha_openai_dive::v1::resources::chat::ChatCompletionParameters;
+
+#[tokio::test]
+async fn index_tts2_generate() -> Result<()> {
+    // RUST_BACKTRACE=1 cargo test -F cuda index_tts2_generate -r -- --nocapture
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let _ = download_index_tts2_need_model(Some(&save_dir)).await?;
+    let model_path = format!("{}/IndexTeam/IndexTTS-2", save_dir);
+    let message = r#"
+    {
+        "model": "index-tts2",
+        "messages": [
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "audio",
+                        "audio_url": 
+                        {
+                            "url": "file:///home/jhq/Videos/voice_01.wav"
+                        }
+                    },              
+                    {
+                        "type": "text", 
+                        "text": "你好啊"
+                    }
+                ]
+            }
+        ]
+    }
+    "#;
+    let mes: ChatCompletionParameters = serde_json::from_str(message)?;
+    let i_start = Instant::now();
+    let mut voxcpm_generate = IndexTTS2Generate::init(&model_path, None, None)?;
+    let i_duration = i_start.elapsed();
+    println!("Time elapsed in load model is: {:?}", i_duration);
+
+    let i_start = Instant::now();
+    let generate = voxcpm_generate.generate(mes)?;
+    let i_duration = i_start.elapsed();
+    println!("Time elapsed in generate is: {:?}", i_duration);
+    Ok(())
+}

tests/weight_test.rs

@@ -1,6 +1,6 @@
 use std::collections::HashMap;
 
-use aha::utils::{find_type_files, get_device};
+use aha::utils::{find_type_files, get_device, read_pth_tensor_info_cycle};
 use anyhow::Result;
 use candle_core::{Device, pickle::read_all_with_key, safetensors};
 use candle_nn::VarBuilder;
@@ -197,3 +197,41 @@ fn qwen3_weight() -> Result<()> {
     println!("model_list: {:?}", model_list);
     Ok(())
 }
+
+#[test]
+fn index_tts2_weight() -> Result<()> {
+    let save_dir: String =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/IndexTeam/IndexTTS-2/", save_dir); 
+    let s2mel_path = model_path+ "/s2mel.pth";
+    // let wac2vec2_path = model_path+ "/wav2vec2bert_stats.pt";
+    // let model_path = format!("{}/iic/speech_campplus_sv_zh-cn_16k-common/", save_dir);
+    // let campplus_path = model_path+ "/campplus_cn_common.bin";
+    // let model_list = find_type_files(&model_path, "safetensors")?;
+    let model_list = vec![s2mel_path];
+    // let mut dict_to_hashmap = HashMap::new();
+    // let mut dtype = candle_core::DType::F32;
+    for m in model_list {
+        // let dict = read_all_with_key(m, Some("state_dict"))?;
+        // let dict = read_all_with_key(m, Some("net"))?;
+        let dict = read_pth_tensor_info_cycle(m, Some("net.cfm"))?;
+        // dtype = dict[0].1.dtype();
+        for (k, v) in dict {
+            // if k.contains("model") {
+            //     println!("key: {}, tensor shape: {:?}", k, v);
+            // }
+            // dict_to_hashmap.insert(k, v);
+            println!("key: {}, tensor shape: {:?}", k, v);
+        }
+    }
+    // let device = Device::Cpu;
+    // let semantic_codec_path = save_dir.to_string() + "/amphion/MaskGCT/semantic_codec/model.safetensors" ;
+    // let model_list = vec![semantic_codec_path];
+    // for m in model_list {
+    //     let weights = safetensors::load(m, &device)?;
+    //     for (key, tensor) in weights.iter() {            
+    //         println!("=== {} === {:?}", key, tensor.shape());
+    //     }
+    // }
+    Ok(())
+}