Add translation example from chinese, french to english (#681)

This adds:
- a VAD node to detect voice within an audio to avoid having too much
noise
- Bump distil-whisper to whisper-turbo
- Use Argotranslate and OPUS-MT for translation
- Add example dataflow within the example folder

## Get started

```bash
cd examples/translation
dora up

dora build dataflow_zh_en_terminal.yml
dora start dataflow_zh_en_terminal.yml --detach

python pretty_print.py
```

examples/translation/.gitignore

@@ -0,0 +1 @@
+*.pt

examples/translation/README.md

@@ -0,0 +1,24 @@
+# Dora argo example
+
+Make sure to have, dora, pip and cargo installed.
+
+```bash
+dora up
+
+## For chinese
+dora build dataflow_zh_en_terminal.yml
+dora start dataflow_zh_en_terminal.yml --detach
+
+python pretty_print.py
+
+dora stop
+
+
+## For chinese
+dora build dataflow_en_zh_terminal.yml
+dora start dataflow_en_zh_terminal.yml --detach
+
+python pretty_print.py
+
+dora stop
+```

examples/translation/dataflow_en_zh.yml

@@ -0,0 +1,45 @@
+nodes:
+  - id: dora-microphone
+    build: pip install -e ../../node-hub/dora-microphone
+    path: dora-microphone
+    outputs:
+      - audio
+
+  - id: dora-vad
+    build: pip install -e ../../node-hub/dora-vad
+    path: dora-vad
+    inputs:
+      audio: dora-microphone/audio
+    outputs:
+      - audio
+
+  - id: dora-distil-whisper
+    build: pip install -e ../../node-hub/dora-distil-whisper
+    path: dora-distil-whisper
+    inputs:
+      input: dora-vad/audio
+    outputs:
+      - text
+    env:
+      TARGET_LANGUAGE: english
+      TRANSLATE: false
+
+  - id: dora-argotranslate
+    build: pip install -e ../../node-hub/dora-argotranslate
+    path: dora-argotranslate
+    inputs:
+      text: dora-distil-whisper/text
+    outputs:
+      - text
+    env:
+      SOURCE_LANGUAGE: en
+      TARGET_LANGUAGE: zh
+
+  - id: dora-rerun
+    build: cargo build -p dora-rerun --release
+    path: dora-rerun
+    inputs:
+      text: dora-argotranslate/text
+    env:
+      IMAGE_WIDTH: 640
+      IMAGE_HEIGHT: 480

examples/translation/dataflow_en_zh_terminal.yml

@@ -0,0 +1,45 @@
+nodes:
+  - id: dora-microphone
+    build: pip install -e ../../node-hub/dora-microphone
+    path: dora-microphone
+    outputs:
+      - audio
+
+  - id: dora-vad
+    build: pip install -e ../../node-hub/dora-vad
+    path: dora-vad
+    inputs:
+      audio: dora-microphone/audio
+    outputs:
+      - audio
+
+  - id: dora-distil-whisper
+    build: pip install -e ../../node-hub/dora-distil-whisper
+    path: dora-distil-whisper
+    inputs:
+      input: dora-vad/audio
+    outputs:
+      - text
+    env:
+      TARGET_LANGUAGE: english
+      TRANSLATE: false
+
+  - id: dora-opus
+    build: pip install -e ../../node-hub/dora-opus
+    path: dora-opus
+    inputs:
+      text: dora-distil-whisper/text
+    outputs:
+      - text
+    env:
+      SOURCE_LANGUAGE: en
+      TARGET_LANGUAGE: zh
+
+  - id: pretty-print
+    path: dynamic
+    inputs:
+      translated_text: dora-opus/text
+      original_text: dora-distil-whisper/text
+    env:
+      IMAGE_WIDTH: 640
+      IMAGE_HEIGHT: 480

examples/translation/dataflow_en_zh_terminal_argo.yml

@@ -0,0 +1,46 @@
+nodes:
+  - id: dora-microphone
+    build: pip install -e ../../node-hub/dora-microphone
+    path: dora-microphone
+    outputs:
+      - audio
+
+  - id: dora-vad
+    build: pip install -e ../../node-hub/dora-vad
+    path: dora-vad
+    inputs:
+      audio: dora-microphone/audio
+    outputs:
+      - audio
+
+  - id: dora-distil-whisper
+    build: pip install -e ../../node-hub/dora-distil-whisper
+    path: dora-distil-whisper
+    inputs:
+      input: dora-vad/audio
+    outputs:
+      - text
+    env:
+      TARGET_LANGUAGE: english
+      TRANSLATE: false
+
+  - id: dora-argotranslate
+    build: pip install -e ../../node-hub/dora-argotranslate
+    path: dora-argotranslate
+    inputs:
+      text: dora-distil-whisper/text
+    outputs:
+      - text
+    env:
+      SOURCE_LANGUAGE: en
+      TARGET_LANGUAGE: zh
+
+  - id: pretty-print
+    build: cargo build -p dora-rerun --release
+    path: dynamic
+    inputs:
+      translated_text: dora-argotranslate/text
+      original_text: dora-distil-whisper/text
+    env:
+      IMAGE_WIDTH: 640
+      IMAGE_HEIGHT: 480

examples/translation/dataflow_fr_en.yml

@@ -0,0 +1,46 @@
+nodes:
+  - id: dora-microphone
+    build: pip install -e ../../node-hub/dora-microphone
+    path: dora-microphone
+    outputs:
+      - audio
+
+  - id: dora-vad
+    build: pip install -e ../../node-hub/dora-vad
+    path: dora-vad
+    inputs:
+      audio: dora-microphone/audio
+    outputs:
+      - audio
+
+  - id: dora-distil-whisper
+    build: pip install -e ../../node-hub/dora-distil-whisper
+    path: dora-distil-whisper
+    inputs:
+      input: dora-vad/audio
+    outputs:
+      - text
+    env:
+      TARGET_LANGUAGE: french
+      TRANSLATE: false
+
+  - id: dora-argotranslate
+    build: pip install -e ../../node-hub/dora-argotranslate
+    path: dora-argotranslate
+    inputs:
+      text: dora-distil-whisper/text
+    outputs:
+      - text
+    env:
+      SOURCE_LANGUAGE: fr
+      TARGET_LANGUAGE: en
+
+  - id: dora-rerun
+    build: cargo build -p dora-rerun --release
+    path: dora-rerun
+    inputs:
+      translated_text: dora-argotranslate/text
+      original_text: dora-distil-whisper/text
+    env:
+      IMAGE_WIDTH: 640
+      IMAGE_HEIGHT: 480

examples/translation/dataflow_zh_en.yml

@@ -0,0 +1,46 @@
+nodes:
+  - id: dora-microphone
+    build: pip install -e ../../node-hub/dora-microphone
+    path: dora-microphone
+    outputs:
+      - audio
+
+  - id: dora-vad
+    build: pip install -e ../../node-hub/dora-vad
+    path: dora-vad
+    inputs:
+      audio: dora-microphone/audio
+    outputs:
+      - audio
+
+  - id: dora-distil-whisper
+    build: pip install -e ../../node-hub/dora-distil-whisper
+    path: dora-distil-whisper
+    inputs:
+      input: dora-vad/audio
+    outputs:
+      - text
+    env:
+      TARGET_LANGUAGE: chinese
+      TRANSLATE: false
+
+  - id: dora-opus
+    build: pip install -e ../../node-hub/dora-opus
+    path: dora-opus
+    inputs:
+      text: dora-distil-whisper/text
+    outputs:
+      - text
+    env:
+      SOURCE_LANGUAGE: zh
+      TARGET_LANGUAGE: en
+
+  - id: plot
+    build: cargo build -p dora-rerun --release
+    path: dora-rerun
+    inputs:
+      translated_text: dora-opus/text
+      original_text: dora-distil-whisper/text
+    env:
+      IMAGE_WIDTH: 640
+      IMAGE_HEIGHT: 480

examples/translation/dataflow_zh_en_terminal.yml

@@ -0,0 +1,42 @@
+nodes:
+  - id: dora-microphone
+    build: pip install -e ../../node-hub/dora-microphone
+    path: dora-microphone
+    outputs:
+      - audio
+
+  - id: dora-vad
+    build: pip install -e ../../node-hub/dora-vad
+    path: dora-vad
+    inputs:
+      audio: dora-microphone/audio
+    outputs:
+      - audio
+
+  - id: dora-distil-whisper
+    build: pip install -e ../../node-hub/dora-distil-whisper
+    path: dora-distil-whisper
+    inputs:
+      input: dora-vad/audio
+    outputs:
+      - text
+    env:
+      TARGET_LANGUAGE: chinese
+      TRANSLATE: false
+
+  - id: dora-opus
+    build: pip install -e ../../node-hub/dora-opus
+    path: dora-opus
+    inputs:
+      text: dora-distil-whisper/text
+    outputs:
+      - text
+    env:
+      SOURCE_LANGUAGE: zh
+      TARGET_LANGUAGE: en
+
+  - id: pretty-print
+    path: dynamic
+    inputs:
+      translated_text: dora-opus/text
+      original_text: dora-distil-whisper/text

examples/translation/pretty_print.py

@@ -0,0 +1,46 @@
+import os
+import shutil
+
+
+def clear_screen():
+    # Clear the screen based on the operating system
+    os.system("cls" if os.name == "nt" else "clear")
+
+
+def print_centered(texts):
+    # Get terminal size
+    terminal_size = shutil.get_terminal_size()
+
+    # Print newlines to move cursor to the middle vertically
+    for k, v in texts.items():
+        print(k)
+        print("\n" * 1)
+        # Calculate horizontal padding and print the centered text
+        for l in v:
+            print(l.center(terminal_size.columns))
+        print("\n" * 1)
+
+
+from dora import Node
+
+node = Node("pretty-print")
+
+previous_texts = {}
+
+clear_screen()
+print("Waiting for speech...")
+for event in node:
+    if event["type"] == "INPUT":
+        # The sentence to be printed
+        sentence = event["value"][0].as_py()
+        if event["id"] not in previous_texts.keys():
+
+            previous_texts[event["id"]] = ["", "", sentence]
+        else:
+            previous_texts[event["id"]] += [sentence]
+            previous_texts[event["id"]] = previous_texts[event["id"]][-3:]
+        # Clear the screen
+        clear_screen()
+
+        # Print the sentence in the middle of the terminal
+        print_centered(previous_texts)

node-hub/dora-argotranslate/README.md

@@ -0,0 +1,2 @@
+# Dora text translation Node using Argo translate
+

node-hub/dora-argotranslate/dora_argotranslate/__init__.py

@@ -0,0 +1,11 @@
+import os
+
+# Define the path to the README file relative to the package directory
+readme_path = os.path.join(os.path.dirname(os.path.dirname(__file__)), "README.md")
+
+# Read the content of the README file
+try:
+    with open(readme_path, "r", encoding="utf-8") as f:
+        __doc__ = f.read()
+except FileNotFoundError:
+    __doc__ = "README file not found."

node-hub/dora-argotranslate/dora_argotranslate/main.py

@@ -0,0 +1,43 @@
+import os
+
+os.environ["ARGOS_DEVICE_TYPE"] = "auto"
+
+from dora import Node
+import pyarrow as pa
+import argostranslate.package
+import argostranslate.translate
+
+from_code = os.getenv("SOURCE_LANGUAGE", "fr")
+to_code = os.getenv("TARGET_LANGUAGE", "en")
+
+# Download and install Argos Translate package
+argostranslate.package.update_package_index()
+available_packages = argostranslate.package.get_available_packages()
+package_to_install = next(
+    filter(
+        lambda x: x.from_code == from_code and x.to_code == to_code, available_packages
+    )
+)
+argostranslate.package.install_from_path(package_to_install.download())
+
+
+def main():
+    node = Node()
+    while True:
+        event = node.next()
+        if event is None:
+            break
+        if event["type"] == "INPUT" and event["id"] == "text":
+            text = event["value"][0].as_py()
+            translatedText = argostranslate.translate.translate(
+                text,
+                from_code,
+                to_code,
+            )
+            print(text, flush=True)
+            print("translated: " + translatedText, flush=True)
+            node.send_output(
+                "text",
+                pa.array([translatedText]),
+                {"language": to_code},
+            )

node-hub/dora-argotranslate/pyproject.toml

@@ -0,0 +1,25 @@
+[tool.poetry]
+name = "dora-argotranslate"
+version = "0.3.6"
+description = "Dora Node for Text translating using Argostranslate"
+readme = "README.md"
+authors = [
+    "Haixuan Xavier Tao <tao.xavier@outlook.com>",
+    "Enzo Le Van <dev@enzo-le-van.fr>",
+    "Félix Huang <felix.huang.net@gmail.com>",
+]
+
+packages = [{ include = "dora_argotranslate" }]
+
+[tool.poetry.dependencies]
+dora-rs = "^0.3.6"
+numpy = "< 2.0.0"
+python = "^3.7"
+argostranslate = "^1.9.6"
+
+[tool.poetry.scripts]
+dora-argotranslate = "dora_argotranslate.main:main"
+
+[build-system]
+requires = ["poetry-core>=1.8.0"]
+build-backend = "poetry.core.masonry.api"

node-hub/dora-argotranslate/tests/test_translate.py

@@ -0,0 +1,9 @@
+import pytest
+
+
+def test_import_main():
+    from dora_argotranslate.main import main
+
+    # Check that everything is working, and catch dora Runtime Exception as we're not running in a dora dataflow.
+    with pytest.raises(RuntimeError):
+        main()

node-hub/dora-distil-whisper/dora_distil_whisper/main.py

@@ -2,34 +2,110 @@ import torch
 from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
 from dora import Node
 import pyarrow as pa
+import os
+from pathlib import Path
 
+DEFAULT_PATH = "openai/whisper-large-v3-turbo"
+TARGET_LANGUAGE = os.getenv("TARGET_LANGUAGE", "chinese")
+TRANSLATE = bool(os.getenv("TRANSLATE", "False"))
+
+
+MODEL_NAME_OR_PATH = os.getenv("MODEL_NAME_OR_PATH", DEFAULT_PATH)
+
+if bool(os.getenv("USE_MODELSCOPE_HUB")) is True:
+    from modelscope import snapshot_download
+
+    if not Path(MODEL_NAME_OR_PATH).exists():
+        MODEL_NAME_OR_PATH = snapshot_download(MODEL_NAME_OR_PATH)
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
 
-model_id = "distil-whisper/distil-large-v3"
 
 model = AutoModelForSpeechSeq2Seq.from_pretrained(
-    model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
+    MODEL_NAME_OR_PATH,
+    torch_dtype=torch_dtype,
+    low_cpu_mem_usage=True,
+    use_safetensors=True,
 )
 model.to(device)
 
-processor = AutoProcessor.from_pretrained(model_id)
+processor = AutoProcessor.from_pretrained(MODEL_NAME_OR_PATH)
 pipe = pipeline(
     "automatic-speech-recognition",
     model=model,
     tokenizer=processor.tokenizer,
     feature_extractor=processor.feature_extractor,
-    max_new_tokens=128,
+    max_new_tokens=400,
     torch_dtype=torch_dtype,
     device=device,
 )
 
+BAD_SENTENCES = [
+    "字幕",
+    "字幕志愿",
+    "中文字幕",
+    "我",
+    "你",
+    "THANK YOU",
+    " Thank you.",
+    " www.microsoft.com",
+    " The",
+    " BANG",
+    " Silence.",
+    " Sous-titrage Société Radio-Canada",
+    " Sous",
+    " Sous-",
+]
+
+
+def cut_repetition(text, min_repeat_length=4, max_repeat_length=50):
+    # Check if the text is primarily Chinese (you may need to adjust this threshold)
+    if sum(1 for char in text if "\u4e00" <= char <= "\u9fff") / len(text) > 0.5:
+        # Chinese text processing
+        for repeat_length in range(
+            min_repeat_length, min(max_repeat_length, len(text) // 2)
+        ):
+            for i in range(len(text) - repeat_length * 2 + 1):
+                chunk1 = text[i : i + repeat_length]
+                chunk2 = text[i + repeat_length : i + repeat_length * 2]
+
+                if chunk1 == chunk2:
+                    return text[: i + repeat_length]
+    else:
+        # Non-Chinese (space-separated) text processing
+        words = text.split()
+        for repeat_length in range(
+            min_repeat_length, min(max_repeat_length, len(words) // 2)
+        ):
+            for i in range(len(words) - repeat_length * 2 + 1):
+                chunk1 = " ".join(words[i : i + repeat_length])
+                chunk2 = " ".join(words[i + repeat_length : i + repeat_length * 2])
+
+                if chunk1 == chunk2:
+                    return " ".join(words[: i + repeat_length])
+
+    return text
+
 
 def main():
     node = Node()
     for event in node:
         if event["type"] == "INPUT":
             audio = event["value"].to_numpy()
-            result = pipe(audio)
-            node.send_output("text", pa.array([result["text"]]))
+            confg = (
+                {"language": TARGET_LANGUAGE, "task": "translate"}
+                if TRANSLATE
+                else {
+                    "language": TARGET_LANGUAGE,
+                }
+            )
+            result = pipe(
+                audio,
+                generate_kwargs=confg,
+            )
+            if result["text"] in BAD_SENTENCES:
+                continue
+            text = cut_repetition(result["text"])
+            print(text, flush=True)
+            node.send_output("text", pa.array([text]), {"language": TARGET_LANGUAGE})

node-hub/dora-distil-whisper/pyproject.toml

@@ -20,6 +20,7 @@ transformers = "^4.0.0"
 accelerate = "^0.29.2"
 torch = "^2.2.0"
 python = "^3.7"
+modelscope = "^1.18.1"
 
 [tool.poetry.scripts]
 dora-distil-whisper = "dora_distil_whisper.main:main"

node-hub/dora-microphone/dora_microphone/main.py

@@ -2,66 +2,31 @@ import sounddevice as sd
 import numpy as np
 import pyarrow as pa
 import time as tm
-from enum import Enum
+import os
 
 from dora import Node
 
-
-class RecordingState(Enum):
-    """Enum for recording states."""
-
-    PENDING = 0
-    RUNNING = 1
-    SILENCE = 2
-
-
-def detect_speech(audio_data, threshold):
-    """Check if the amplitude of the audio signal exceeds the threshold."""
-    return np.any(np.abs(audio_data) > threshold)
+MAX_DURATION = float(os.getenv("MAX_DURATION", "0.1"))
+SAMPLE_RATE = int(os.getenv("SAMPLE_RATE", "16000"))
 
 
 def main():
-    # Parameters
-    threshold = 500  # Threshold for detecting speech (adjust this as needed)
-    SAMPLE_RATE = 16000
-    silence_duration = 0.5  # Duration of silence before stopping the recording
-
     # Initialize buffer and recording flag
     buffer = []
-    state = RecordingState.PENDING
-    silence_start_time = tm.time()
     start_recording_time = tm.time()
-    max_duration = 20
     node = Node()
 
     # pylint: disable=unused-argument
     def callback(indata, frames, time, status):
-        nonlocal buffer, state, silence_start_time, node, max_duration, start_recording_time
+        nonlocal buffer, node, start_recording_time
 
-        is_speaking = detect_speech(indata[:, 0], threshold)
-        if is_speaking:
-            if state == RecordingState.PENDING:
-                buffer = []
-                state = RecordingState.RUNNING
-                start_recording_time = tm.time()
-            buffer.extend(indata[:, 0])
-        elif not is_speaking and state == RecordingState.RUNNING:
-            silence_start_time = tm.time()  # Reset silence timer
-            buffer.extend(indata[:, 0])
-            state = RecordingState.SILENCE
-        elif (
-            state == RecordingState.RUNNING or state == RecordingState.SILENCE
-        ) and tm.time() - start_recording_time > max_duration:
+        if tm.time() - start_recording_time > MAX_DURATION:
             audio_data = np.array(buffer).ravel().astype(np.float32) / 32768.0
             node.send_output("audio", pa.array(audio_data))
-            state = RecordingState.PENDING
-        elif not is_speaking and state == RecordingState.SILENCE:
-            if tm.time() - silence_start_time > silence_duration:
-                audio_data = np.array(buffer).ravel().astype(np.float32) / 32768.0
-                node.send_output("audio", pa.array(audio_data))
-                state = RecordingState.PENDING
-            else:
-                buffer.extend(indata[:, 0])
+            buffer = []
+            start_recording_time = tm.time()
+        else:
+            buffer.extend(indata[:, 0])
 
     # Start recording
     with sd.InputStream(

node-hub/dora-opus/README.md

@@ -0,0 +1 @@
+# Dora text translation Node using OPUS MT

node-hub/dora-opus/dora_opus/__init__.py

@@ -0,0 +1,11 @@
+import os
+
+# Define the path to the README file relative to the package directory
+readme_path = os.path.join(os.path.dirname(os.path.dirname(__file__)), "README.md")
+
+# Read the content of the README file
+try:
+    with open(readme_path, "r", encoding="utf-8") as f:
+        __doc__ = f.read()
+except FileNotFoundError:
+    __doc__ = "README file not found."

node-hub/dora-opus/dora_opus/main.py

@@ -0,0 +1,78 @@
+import os
+from pathlib import Path
+from dora import Node
+import pyarrow as pa
+import numpy as np
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+
+from_code = os.getenv("SOURCE_LANGUAGE", "zh")
+to_code = os.getenv("TARGET_LANGUAGE", "en")
+DEFAULT_PATH = f"Helsinki-NLP/opus-mt-{from_code}-{to_code}"
+
+
+MODEL_NAME_OR_PATH = os.getenv("MODEL_NAME_OR_PATH", DEFAULT_PATH)
+
+if bool(os.getenv("USE_MODELSCOPE_HUB")) is True:
+    from modelscope import snapshot_download
+
+    if not Path(MODEL_NAME_OR_PATH).exists():
+        MODEL_NAME_OR_PATH = snapshot_download(MODEL_NAME_OR_PATH)
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME_OR_PATH)
+
+model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME_OR_PATH)
+
+
+def cut_repetition(text, min_repeat_length=4, max_repeat_length=50):
+    # Check if the text is primarily Chinese (you may need to adjust this threshold)
+    if sum(1 for char in text if "\u4e00" <= char <= "\u9fff") / len(text) > 0.5:
+        # Chinese text processing
+        for repeat_length in range(
+            min_repeat_length, min(max_repeat_length, len(text) // 2)
+        ):
+            for i in range(len(text) - repeat_length * 2 + 1):
+                chunk1 = text[i : i + repeat_length]
+                chunk2 = text[i + repeat_length : i + repeat_length * 2]
+
+                if chunk1 == chunk2:
+                    return text[: i + repeat_length]
+    else:
+        # Non-Chinese (space-separated) text processing
+        words = text.split()
+        for repeat_length in range(
+            min_repeat_length, min(max_repeat_length, len(words) // 2)
+        ):
+            for i in range(len(words) - repeat_length * 2 + 1):
+                chunk1 = " ".join(words[i : i + repeat_length])
+                chunk2 = " ".join(words[i + repeat_length : i + repeat_length * 2])
+
+                if chunk1 == chunk2:
+                    return " ".join(words[: i + repeat_length])
+
+    return text
+
+
+def main():
+    node = Node()
+    while True:
+        event = node.next()
+        if event is None:
+            break
+        if event["type"] == "INPUT" and event["id"] == "text":
+            text = [str(event["value"][0].as_py())]
+            translated = (
+                model.generate(**tokenizer(text, return_tensors="pt", padding=True))
+                .to("cpu")
+                .detach()
+                .numpy()
+                .ravel()
+            )
+
+            array = np.array(tokenizer.decode(translated, skip_special_tokens=True))
+            array = np.array(array).ravel()
+            array = [cut_repetition(array[0])]
+            node.send_output(
+                "text",
+                pa.array(array),
+                {"language": to_code},
+            )

node-hub/dora-opus/pyproject.toml

@@ -0,0 +1,26 @@
+[tool.poetry]
+name = "dora-opus"
+version = "0.3.6"
+description = "Dora Node for Text translating using Opus"
+readme = "README.md"
+authors = [
+    "Haixuan Xavier Tao <tao.xavier@outlook.com>",
+    "Enzo Le Van <dev@enzo-le-van.fr>",
+    "Félix Huang <felix.huang.net@gmail.com>",
+]
+
+packages = [{ include = "dora_opus" }]
+
+[tool.poetry.dependencies]
+dora-rs = "^0.3.6"
+numpy = "< 2.0.0"
+python = "^3.7"
+transformers = "^4.45"
+modelscope = "^1.18.1"
+
+[tool.poetry.scripts]
+dora-opus = "dora_opus.main:main"
+
+[build-system]
+requires = ["poetry-core>=1.8.0"]
+build-backend = "poetry.core.masonry.api"

node-hub/dora-opus/tests/test_translate.py

@@ -0,0 +1,9 @@
+import pytest
+
+
+def test_import_main():
+    from dora_opus.main import main
+
+    # Check that everything is working, and catch dora Runtime Exception as we're not running in a dora dataflow.
+    with pytest.raises(RuntimeError):
+        main()

node-hub/dora-parler/dora_parler/main.py

@@ -1,6 +1,7 @@
 from threading import Thread
 from dora import Node
-
+import os
+from pathlib import Path
 import numpy as np
 import torch
 import time
@@ -18,16 +19,25 @@ from transformers import (
 device = "cuda:0"  # if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
 torch_dtype = torch.float16 if device != "cpu" else torch.float32
 
-repo_id = "ylacombe/parler-tts-mini-jenny-30H"
+DEFAULT_PATH = "ylacombe/parler-tts-mini-jenny-30H"
+
+
+MODEL_NAME_OR_PATH = os.getenv("MODEL_NAME_OR_PATH", DEFAULT_PATH)
+
+if bool(os.getenv("USE_MODELSCOPE_HUB")) is True:
+    from modelscope import snapshot_download
+
+    if not Path(MODEL_NAME_OR_PATH).exists():
+        MODEL_NAME_OR_PATH = snapshot_download(MODEL_NAME_OR_PATH)
 
 model = ParlerTTSForConditionalGeneration.from_pretrained(
-    repo_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True
+    MODEL_NAME_OR_PATH, torch_dtype=torch_dtype, low_cpu_mem_usage=True
 ).to(device)
 model.generation_config.cache_implementation = "static"
 model.forward = torch.compile(model.forward, mode="default")
 
-tokenizer = AutoTokenizer.from_pretrained(repo_id)
-feature_extractor = AutoFeatureExtractor.from_pretrained(repo_id)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME_OR_PATH)
+feature_extractor = AutoFeatureExtractor.from_pretrained(MODEL_NAME_OR_PATH)
 
 SAMPLE_RATE = feature_extractor.sampling_rate
 SEED = 42
@@ -59,6 +69,7 @@ def play_audio(audio_array):
 
 class InterruptStoppingCriteria(StoppingCriteria):
     def __init__(self):
+        super().__init__()
         self.stop_signal = False
 
     def __call__(

node-hub/dora-parler/pyproject.toml

@@ -20,6 +20,7 @@ torchaudio = "^2.2.2"
 sentencepiece = "^0.1.99"
 python = "^3.7"
 pyaudio = "^0.2.14"
+modelscope = "^1.18.1"
 
 
 [tool.poetry.scripts]

node-hub/dora-qwenvl/dora_qwenvl/main.py

@@ -12,7 +12,7 @@ DEFAULT_PATH = "Qwen/Qwen2-VL-2B-Instruct"
 
 MODEL_NAME_OR_PATH = os.getenv("MODEL_NAME_OR_PATH", DEFAULT_PATH)
 
-if bool(os.getenv("MODELSCOPE")) is True:
+if bool(os.getenv("USE_MODELSCOPE_HUB")) is True:
     from modelscope import snapshot_download
 
     if not Path(MODEL_NAME_OR_PATH).exists():

node-hub/dora-vad/README.md

@@ -0,0 +1,3 @@
+# Speech Activity Detection(VAD)
+
+This is using Silero VAD.

node-hub/dora-vad/dora_vad/__init__.py

@@ -0,0 +1,11 @@
+import os
+
+# Define the path to the README file relative to the package directory
+readme_path = os.path.join(os.path.dirname(os.path.dirname(__file__)), "README.md")
+
+# Read the content of the README file
+try:
+    with open(readme_path, "r", encoding="utf-8") as f:
+        __doc__ = f.read()
+except FileNotFoundError:
+    __doc__ = "README file not found."

node-hub/dora-vad/dora_vad/main.py

@@ -0,0 +1,53 @@
+from dora import Node
+import pyarrow as pa
+import numpy as np
+import os
+from silero_vad import load_silero_vad, get_speech_timestamps
+import torch
+
+model = load_silero_vad()
+MIN_SILENCE_DURATION_MS = int(os.getenv("MIN_SILENCE_DURATION_MS", "200"))
+MIN_SPEECH_DURATION_MS = int(os.getenv("MIN_SPEECH_DURATION_MS", "1000"))
+
+MIN_AUDIO_SAMPLING_DURAION_S = int(os.getenv("MAX_AUDIO_DURATION_S", "20"))
+MAX_AUDIO_DURAION_S = int(os.getenv("MAX_AUDIO_DURATION_S", "75"))
+
+
+def main():
+    node = Node()
+    last_audios = []
+    while True:
+        event = node.next()
+        if event is None:
+            break
+        if event["type"] == "INPUT" and event["id"] == "audio":
+            audio = event["value"].to_numpy()
+            last_audios += [audio]
+            last_audios = last_audios[-100:]
+            audio = np.concatenate(last_audios)
+            speech_timestamps = get_speech_timestamps(
+                torch.from_numpy(audio),
+                model,
+                threshold=0.2,
+                min_speech_duration_ms=MIN_SPEECH_DURATION_MS,
+                min_silence_duration_ms=MIN_SILENCE_DURATION_MS,
+            )
+
+            # Check ig there is timestamp
+            if (
+                len(speech_timestamps) > 0
+                and len(last_audios) > MIN_AUDIO_SAMPLING_DURAION_S
+            ):
+
+                # Check if the audio is not cut at the end. And only return if there is a long time spent
+                if speech_timestamps[-1]["end"] == len(audio):
+                    continue
+                else:
+                    audio = audio[0 : speech_timestamps[-1]["end"]]
+                    node.send_output("audio", pa.array(audio))
+                    last_audios = [audio[speech_timestamps[-1]["end"] :]]
+
+            # If there is no sound for too long return the audio
+            elif len(last_audios) > 75:
+                node.send_output("audio", pa.array(audio))
+                last_audios = []

node-hub/dora-vad/pyproject.toml

@@ -0,0 +1,26 @@
+[tool.poetry]
+name = "dora-vad"
+version = "0.3.6"
+description = "Dora Node for Text translating using Argostranslate"
+readme = "README.md"
+authors = [
+    "Haixuan Xavier Tao <tao.xavier@outlook.com>",
+    "Enzo Le Van <dev@enzo-le-van.fr>",
+    "Félix Huang <felix.huang.net@gmail.com>",
+]
+
+packages = [{ include = "dora_vad" }]
+
+[tool.poetry.dependencies]
+dora-rs = "^0.3.6"
+numpy = "< 2.0.0"
+python = "^3.7"
+silero-vad = "^5.1"
+
+
+[tool.poetry.scripts]
+dora-vad = "dora_vad.main:main"
+
+[build-system]
+requires = ["poetry-core>=1.8.0"]
+build-backend = "poetry.core.masonry.api"

node-hub/dora-vad/tests/test_translate.py

@@ -0,0 +1,9 @@
+import pytest
+
+
+def test_import_main():
+    from dora_vad.main import main
+
+    # Check that everything is working, and catch dora Runtime Exception as we're not running in a dora dataflow.
+    with pytest.raises(RuntimeError):
+        main()