init commit for all files

app.py

@@ -0,0 +1,167 @@
+import gradio as gr
+import subprocess, json, os, io, tempfile
+from faster_whisper import WhisperModel
+from ollama import Client as OllamaClient
+
+# ---- CONFIG ----
+LLM_MODEL = "llama3.2:3b"      # or "mistral:7b", "qwen2.5:3b"
+WHISPER_SIZE = "small"         # "base", "small", "medium"
+USE_SILERO = True              # set False to use Coqui XTTS v2
+
+import os
+USE_REMOTE_OLLAMA = bool(os.getenv("OLLAMA_HOST"))
+
+if not USE_REMOTE_OLLAMA:
+    # Transformers fallback for Spaces (CPU-friendly small instruct model)
+    from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+    HF_CHAT_MODEL = os.getenv("HF_CHAT_MODEL", "google/gemma-2-2b-it")  # small instruct model that runs on CPU
+    _tok = AutoTokenizer.from_pretrained(HF_CHAT_MODEL)
+    _mdl = AutoModelForCausalLM.from_pretrained(HF_CHAT_MODEL, torch_dtype="auto", device_map="auto")
+    gen = pipeline("text-generation", model=_mdl, tokenizer=_tok, max_new_tokens=256)
+
+
+# ---- STT (faster-whisper) ----
+# Run on GPU if available: compute_type="float16", device="cuda"
+stt_model = WhisperModel(WHISPER_SIZE, device="cuda" if os.environ.get("CUDA_VISIBLE_DEVICES") else "cpu",
+                         compute_type="float16" if os.environ.get("CUDA_VISIBLE_DEVICES") else "int8")
+
+def speech_to_text(audio_path: str) -> str:
+    segments, info = stt_model.transcribe(audio_path, beam_size=1, vad_filter=True)
+    text = "".join(seg.text for seg in segments).strip()
+    return text
+
+# ---- LLM (Ollama) ----
+ollama = OllamaClient(host="http://127.0.0.1:11434")
+
+SYSTEM_PROMPT = """You are a friendly conversational English coach and voice assistant.
+- First, understand the user's utterance.
+- If there are mistakes (grammar/word choice/tense), provide a brief corrected sentence first, prefixed with "Correction:".
+- In 1 short line, explain the key fix, prefixed with "Why:".
+- Then continue the conversation naturally in one or two sentences.
+- Be concise, supportive, and avoid long lectures.
+Format:
+Correction: <corrected sentence or "None">
+Why: <very brief reason, or "N/A">
+Reply: <your friendly response to keep the conversation going>"""
+
+def chat_with_llm(history_messages, user_text):
+    if USE_REMOTE_OLLAMA:
+        messages = [{"role": "system", "content": SYSTEM_PROMPT}]
+        for m in (history_messages or []):
+            if m.get("role") in ("user", "assistant") and m.get("content"):
+                messages.append({"role": m["role"], "content": m["content"]})
+        messages.append({"role": "user", "content": user_text})
+        resp = ollama.chat(model=LLM_MODEL, messages=messages)
+        return resp["message"]["content"]
+    else:
+        # Simple prompt stitching for the fallback pipeline
+        history_text = "\n".join(
+            [f"User: {m['content']}" if m["role"]=="user" else f"Assistant: {m['content']}"
+             for m in (history_messages or [])]
+        )
+        prompt = f"{SYSTEM_PROMPT}\n{history_text}\nUser: {user_text}\nAssistant:"
+        out = gen(prompt)[0]["generated_text"]
+        # Return only the new assistant chunk after the prompt
+        return out.split("Assistant:", 1)[-1].strip()
+
+
+
+# ---- TTS ----
+def tts_silero(text: str) -> str:
+    """
+    Return path to a WAV file synthesized by Silero (CPU-friendly).
+    Works across recent torch.hub return signatures.
+    """
+    import torch, tempfile
+    import soundfile as sf
+
+    # Newer torch.hub supports "trust_repo"; set to True or 'check'
+    obj = torch.hub.load(
+        repo_or_dir="snakers4/silero-models",
+        model="silero_tts",
+        language="en",
+        speaker="v3_en",
+        trust_repo=True  # or 'check' to be prompted the first time
+    )
+
+    # Handle both cases: either a single model, or a (model, something) tuple
+    model = obj[0] if isinstance(obj, (list, tuple)) else obj
+
+    sample_rate = 48000
+    speaker = "en_0"  # valid default voice in v3_en pack
+    audio = model.apply_tts(text=text, speaker=speaker, sample_rate=sample_rate)
+
+    out_wav = tempfile.mktemp(suffix=".wav")
+    sf.write(out_wav, audio, sample_rate)
+    return out_wav
+
+
+def tts_coqui_xtts(text: str) -> str:
+    """
+    Returns path to a WAV file synthesized by Coqui XTTS v2 (higher quality; GPU-friendly).
+    """
+    from TTS.api import TTS
+    tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2")
+    out_wav = tempfile.mktemp(suffix=".wav")
+    tts.tts_to_file(text=text, file_path=out_wav, speaker="female-en-5", language="en")
+    return out_wav
+
+def text_to_speech(text: str) -> str:
+    if USE_SILERO:
+        return tts_silero(text)
+    else:
+        return tts_coqui_xtts(text)
+
+# ---- Gradio pipeline ----
+def pipeline(audio, history):
+    # audio is (sample_rate, np.array) OR a filepath (depends on Gradio version)
+    # Normalize to a temp wav file
+    if audio is None:
+        return history, None, "Please speak something."
+
+    if isinstance(audio, tuple):
+        # (sr, data) -> write wav
+        import soundfile as sf, numpy as np, tempfile
+        sr, data = audio
+        tmp_in = tempfile.mktemp(suffix=".wav")
+        sf.write(tmp_in, data.astype("float32"), sr)
+        audio_path = tmp_in
+    else:
+        audio_path = audio  # path already
+
+    user_text = speech_to_text(audio_path)
+    if not user_text:
+        return history, None, "Didn't catch that—could you repeat?"
+
+    reply = chat_with_llm(history, user_text)
+
+    # Extract the "Reply:" line for TTS; speak only the conversational reply
+    speak_text = reply
+    for tag in ["Reply:", "Correction:", "Why:"]:
+        # Try to find "Reply:" block
+        if "Reply:" in reply:
+            speak_text = reply.split("Reply:", 1)[1].strip()
+            break
+
+    wav_path = text_to_speech(speak_text)
+    updated = (history or []) + [
+        {"role": "user", "content": user_text},
+        {"role": "assistant", "content": reply},
+    ]
+    return updated, wav_path, ""
+
+with gr.Blocks(title="Voice Coach") as demo:
+    gr.Markdown("## 🎙️ Interactive Voice Chat (with on-the-fly sentence correction)")
+    with gr.Row():
+        audio_in = gr.Audio(sources=["microphone"], type="filepath", label="Speak")
+        audio_out = gr.Audio(label="Assistant (TTS)", autoplay=True)
+    chatbox = gr.Chatbot(type="messages", height=300)
+    status = gr.Markdown()
+    btn = gr.Button("Send")
+
+    # Use continuous recording or press "Send" after recording
+    audio_in.change(pipeline, inputs=[audio_in, chatbox], outputs=[chatbox, audio_out, status])
+    btn.click(pipeline, inputs=[audio_in, chatbox], outputs=[chatbox, audio_out, status])
+
+if __name__ == "__main__":
+    demo.launch(share=True)

apt.txt

@@ -0,0 +1 @@
+ffmpeg

requirements.txt

@@ -0,0 +1,94 @@
+aiofiles==23.2.1
+annotated-doc==0.0.3
+annotated-types==0.7.0
+antlr4-python3-runtime==4.9.3
+anyio==4.11.0
+av==16.0.1
+Brotli==1.1.0
+certifi==2025.10.5
+cffi==2.0.0
+charset-normalizer==3.4.4
+click==8.3.0
+colorama==0.4.6
+coloredlogs==15.0.1
+contourpy==1.3.3
+ctranslate2==4.6.0
+cycler==0.12.1
+fastapi==0.120.3
+faster-whisper==1.2.0
+ffmpy==0.6.4
+filelock==3.19.1
+flatbuffers==25.9.23
+fonttools==4.60.1
+fsspec==2025.9.0
+gradio==5.49.1
+gradio_client==1.13.3
+groovy==0.1.2
+h11==0.16.0
+hf-xet==1.2.0
+httpcore==1.0.9
+httpx==0.28.1
+huggingface-hub==1.0.1
+humanfriendly==10.0
+idna==3.11
+importlib_resources==6.5.2
+Jinja2==3.1.6
+kiwisolver==1.4.9
+markdown-it-py==4.0.0
+MarkupSafe==2.1.5
+matplotlib==3.10.7
+mdurl==0.1.2
+mpmath==1.3.0
+networkx==3.5
+numpy==2.3.4
+ollama==0.6.0
+omegaconf==2.3.0
+onnxruntime==1.23.2
+orjson==3.11.4
+packaging==25.0
+pandas==2.3.3
+pillow==10.4.0
+protobuf==6.33.0
+pycparser==2.23
+pydantic==2.11.10
+pydantic_core==2.33.2
+pydub==0.25.1
+Pygments==2.19.2
+pyparsing==3.2.5
+pyreadline3==3.5.4
+python-dateutil==2.9.0.post0
+python-multipart==0.0.20
+pytz==2025.2
+PyYAML==6.0.3
+requests==2.32.5
+rich==14.2.0
+ruff==0.14.3
+safehttpx==0.1.7
+semantic-version==2.10.0
+shellingham==1.5.4
+silero==0.5.1
+six==1.17.0
+sniffio==1.3.1
+sounddevice==0.5.3
+soundfile==0.13.1
+starlette==0.49.1
+sympy==1.14.0
+tokenizers==0.22.1
+tomlkit==0.12.0
+torch==2.9.0+cu128
+torchaudio==2.9.0
+tqdm==4.67.1
+typer==0.20.0
+typer-slim==0.20.0
+typing-inspection==0.4.2
+typing_extensions==4.15.0
+tzdata==2025.2
+urllib3==2.5.0
+uvicorn==0.38.0
+websockets==15.0.1
+transformers==4.44.2
+accelerate
+safetensors
+sentencepiece
+protobuf
+einops

runtime.txt

@@ -0,0 +1 @@
+python-3.11