From e409c4a19b911f203de69f043987e9f37d78b0c8 Mon Sep 17 00:00:00 2001
From: Yrom <yrom@qq.com>
Date: Thu, 11 Sep 2025 16:55:18 +0800
Subject: [PATCH 1/3] fix(infer_v2): Correct the import path of BigVGAN's
 custom cuda kernel

---
 indextts/infer_v2.py | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/indextts/infer_v2.py b/indextts/infer_v2.py
index 71c36e1..ab5ef73 100644
--- a/indextts/infer_v2.py
+++ b/indextts/infer_v2.py
@@ -100,12 +100,12 @@ class IndexTTS2:
         if self.use_cuda_kernel:
             # preload the CUDA kernel for BigVGAN
             try:
-                from indextts.BigVGAN.alias_free_activation.cuda import load
+                from indextts.s2mel.modules.bigvgan.alias_free_activation.cuda import activation1d
 
-                anti_alias_activation_cuda = load.load()
-                print(">> Preload custom CUDA kernel for BigVGAN", anti_alias_activation_cuda)
-            except:
+                print(">> Preload custom CUDA kernel for BigVGAN", activation1d.anti_alias_activation_cuda)
+            except Exception as e:
                 print(">> Failed to load custom CUDA kernel for BigVGAN. Falling back to torch.")
+                print(f"{e!r}")
                 self.use_cuda_kernel = False
 
         self.extract_features = SeamlessM4TFeatureExtractor.from_pretrained("facebook/w2v-bert-2.0")

From 0828dcb098247760ba17ba5ad8a5b6b5cb460f95 Mon Sep 17 00:00:00 2001
From: nanaoto <nanaoto@foxmail.com>
Date: Fri, 12 Sep 2025 16:45:37 +0800
Subject: [PATCH 2/3] =?UTF-8?q?feat:=20=E8=A3=81=E5=89=AA=E8=BF=87?=
 =?UTF-8?q?=E9=95=BF=E7=9A=84=E8=BE=93=E5=85=A5=E9=9F=B3=E9=A2=91=E8=87=B3?=
 =?UTF-8?q?15s,=E5=87=8F=E5=B0=91=E7=88=86=E5=86=85=E5=AD=98=E5=92=8C?=
 =?UTF-8?q?=E6=98=BE=E5=AD=98?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 indextts/infer_v2.py | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)

diff --git a/indextts/infer_v2.py b/indextts/infer_v2.py
index 71c36e1..972339c 100644
--- a/indextts/infer_v2.py
+++ b/indextts/infer_v2.py
@@ -292,6 +292,20 @@ class IndexTTS2:
         if self.gr_progress is not None:
             self.gr_progress(value, desc=desc)
 
+    def _load_and_cut_audio(self,audio_path,max_audio_length_seconds,verbose=False,sr=None):
+        if not sr:
+            audio, sr = librosa.load(audio_path)
+        else:
+            audio, _ = librosa.load(audio_path,sr=sr)
+        audio = torch.tensor(audio).unsqueeze(0)
+        max_audio_samples = int(max_audio_length_seconds * sr)
+
+        if audio.shape[1] > max_audio_samples:
+            if verbose:
+                print(f"Audio too long ({audio.shape[1]} samples), truncating to {max_audio_samples} samples")
+            audio = audio[:, :max_audio_samples]
+        return audio, sr
+
     # 原始推理模式
     def infer(self, spk_audio_prompt, text, output_path,
               emo_audio_prompt=None, emo_alpha=1.0,
@@ -340,8 +354,7 @@ class IndexTTS2:
 
         # 如果参考音频改变了，才需要重新生成, 提升速度
         if self.cache_spk_cond is None or self.cache_spk_audio_prompt != spk_audio_prompt:
-            audio, sr = librosa.load(spk_audio_prompt)
-            audio = torch.tensor(audio).unsqueeze(0)
+            audio,sr = self._load_and_cut_audio(spk_audio_prompt,15,verbose)
             audio_22k = torchaudio.transforms.Resample(sr, 22050)(audio)
             audio_16k = torchaudio.transforms.Resample(sr, 16000)(audio)
 
@@ -392,7 +405,7 @@ class IndexTTS2:
             emovec_mat = emovec_mat.unsqueeze(0)
 
         if self.cache_emo_cond is None or self.cache_emo_audio_prompt != emo_audio_prompt:
-            emo_audio, _ = librosa.load(emo_audio_prompt, sr=16000)
+            emo_audio, _ = self._load_and_cut_audio(emo_audio_prompt,15,verbose,sr=16000)
             emo_inputs = self.extract_features(emo_audio, sampling_rate=16000, return_tensors="pt")
             emo_input_features = emo_inputs["input_features"]
             emo_attention_mask = emo_inputs["attention_mask"]

From a05850286588600c632675bb128f628dc58b9070 Mon Sep 17 00:00:00 2001
From: nanaoto <nanaoto@foxmail.com>
Date: Mon, 15 Sep 2025 17:47:08 +0800
Subject: [PATCH 3/3] Add Docker publish workflow configuration

---
 .github/workflows/docker-publish.yml | 44 ++++++++++++++++++++++++++++
 1 file changed, 44 insertions(+)
 create mode 100644 .github/workflows/docker-publish.yml

diff --git a/.github/workflows/docker-publish.yml b/.github/workflows/docker-publish.yml
new file mode 100644
index 0000000..f3e2572
--- /dev/null
+++ b/.github/workflows/docker-publish.yml
@@ -0,0 +1,44 @@
+name: Build and Publish Docker Image
+on:
+  workflow_dispatch:
+
+jobs:
+  build-amd64:
+    runs-on: ubuntu-22.04
+    strategy:
+      matrix:
+        include:
+          - cuda_version: 11.8
+            torch_version: 2.4.1
+            tag_prefix: pytorch2.4.1-cuda11.8
+          - cuda_version: 12.8
+            torch_version: 2.8.0
+            tag_prefix: pytorch2.8.0-cuda12.8
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Extract Docker Meta
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: nanaoto/index-tts
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Build Docker Image
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          file: ./Dockerfile
+          push: false
+          platforms: linux/amd64
+          build-args: |
+            CUDA_VERSION=${{ matrix.cuda_version }}
+            TORCH_VERSION=${{ matrix.torch_version }}
+          tags: |
+            nanaoto/index-tts:${{ matrix.tag_prefix }}-${{ steps.meta.outputs.tags }}-amd64
+            nanaoto/index-tts:latest-${{ matrix.tag_prefix }}-amd64
+            
+