Upload 11 files

Dockerfile

@@ -0,0 +1,19 @@
+FROM python:3.10-slim
+
+# 1. Installation de libsndfile (obligatoire pour lire l'audio)
+RUN apt-get update && apt-get install -y libsndfile1 && rm -rf /var/lib/apt/lists/*
+
+# 2. Configuration du dossier de travail
+WORKDIR /code
+
+# 3. Installation des dépendances Python
+COPY ./requirements.txt /code/requirements.txt
+RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
+
+# 4. Copie du modèle et du code
+# Important : on copie le dossier model_fr_onnx entier
+COPY ./model_fr_onnx /code/model_fr_onnx
+COPY ./app.py /code/app.py
+
+# 5. Lancement du serveur (Port 7860 imposé par Hugging Face)
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

app.py

@@ -0,0 +1,79 @@
+from fastapi import FastAPI, File, UploadFile, Form, HTTPException
+from optimum.onnxruntime import ORTModelForCTC
+from transformers import Wav2Vec2Processor
+import torch
+import librosa
+import numpy as np
+import io
+from contextlib import asynccontextmanager
+
+# --- CONFIGURATION ---
+# Le dossier sera copié par le Dockerfile au même niveau que app.py
+ONNX_MODEL_DIR = "model_fr_onnx"
+ONNX_FILENAME = "model_quantized.onnx"
+
+ai_context = {}
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    print("🚀 Chargement du modèle Français ONNX...")
+    try:
+        ai_context["processor"] = Wav2Vec2Processor.from_pretrained(ONNX_MODEL_DIR)
+        ai_context["model"] = ORTModelForCTC.from_pretrained(ONNX_MODEL_DIR, file_name=ONNX_FILENAME)
+        ai_context["vocab"] = ai_context["processor"].tokenizer.get_vocab()
+        print("✅ Modèle chargé.")
+    except Exception as e:
+        print(f"❌ Erreur critique : {e}")
+    yield
+    ai_context.clear()
+
+app = FastAPI(lifespan=lifespan)
+
+@app.get("/")
+def home():
+    return {"status": "API is running", "help": "POST /transcribe to use"}
+
+@app.post("/transcribe")
+async def transcribe(
+    file: UploadFile = File(...), 
+    allowed_phones: str = Form(...) 
+):
+    if "model" not in ai_context:
+        raise HTTPException(status_code=500, detail="Model not loaded")
+    
+    # Lecture Audio
+    audio_content = await file.read()
+    try:
+        speech, _ = librosa.load(io.BytesIO(audio_content), sr=16000)
+    except:
+         # Fallback si librosa n'aime pas le format direct, on peut essayer soundfile
+         import soundfile as sf
+         speech, _ = sf.read(io.BytesIO(audio_content))
+         if len(speech.shape) > 1: speech = speech[:, 0] # Stereo to mono
+         if _ != 16000: speech = librosa.resample(speech, orig_sr=_, target_sr=16000)
+
+    # Inférence
+    processor = ai_context["processor"]
+    inputs = processor(speech, sampling_rate=16000, return_tensors="pt", padding=True)
+    
+    logits = ai_context["model"](inputs.input_values).logits
+    
+    # Masquage Dynamique
+    user_allowed = [p.strip() for p in allowed_phones.split(',')]
+    technical_tokens = ["|", "[PAD]", "<s>", "</s>", "<pad>", "<unk>", "[UNK]"]
+    full_allowed = set(user_allowed + technical_tokens)
+    
+    vocab = ai_context["vocab"]
+    mask = torch.ones(logits.shape[-1], dtype=torch.bool)
+    
+    allowed_indices = [vocab[t] for t in full_allowed if t in vocab]
+    if allowed_indices:
+        mask[allowed_indices] = False
+        
+    logits[:, :, mask] = -float('inf')
+
+    # Décodage
+    predicted_ids = torch.argmax(logits, dim=-1)
+    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
+    
+    return {"ipa": transcription}

model_fr_onnx/added_tokens.json

@@ -0,0 +1,4 @@
+{
+  "</s>": 60,
+  "<s>": 59
+}

model_fr_onnx/config.json

@@ -0,0 +1,118 @@
+{
+  "_name_or_path": "model_fr_temp",
+  "activation_dropout": 0.0,
+  "adapter_attn_dim": null,
+  "adapter_kernel_size": 3,
+  "adapter_stride": 2,
+  "add_adapter": false,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 59,
+  "classifier_proj_size": 256,
+  "codevector_dim": 256,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": false,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": false,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": false,
+  "eos_token_id": 60,
+  "feat_extract_activation": "gelu",
+  "feat_extract_norm": "group",
+  "feat_proj_dropout": 0.1,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0,
+  "freeze_feat_extract_train": true,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.05,
+  "mask_channel_length": 10,
+  "mask_channel_min_space": 1,
+  "mask_channel_other": 0.0,
+  "mask_channel_prob": 0.0,
+  "mask_channel_selection": "static",
+  "mask_feature_length": 64,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.1024,
+  "mask_time_length": 10,
+  "mask_time_min_masks": 2,
+  "mask_time_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.5,
+  "mask_time_selection": "static",
+  "model_type": "wav2vec2",
+  "no_mask_channel_overlap": false,
+  "no_mask_time_overlap": false,
+  "num_adapter_layers": 3,
+  "num_attention_heads": 12,
+  "num_codevector_groups": 2,
+  "num_codevectors_per_group": 320,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 12,
+  "num_negatives": 100,
+  "output_hidden_size": 768,
+  "pad_token_id": 58,
+  "proj_codevector_dim": 256,
+  "tdnn_dilation": [
+    1,
+    2,
+    3,
+    1,
+    1
+  ],
+  "tdnn_dim": [
+    512,
+    512,
+    512,
+    512,
+    1500
+  ],
+  "tdnn_kernel": [
+    5,
+    3,
+    3,
+    1,
+    1
+  ],
+  "torch_dtype": "float32",
+  "transformers_version": "4.38.2",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 61,
+  "xvector_output_dim": 512
+}

model_fr_onnx/model_quantized.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:790f5addf53746f55c42493dab3b042af5fae65c35cfd7871dd6edaac25bd4d1
+size 95349071

model_fr_onnx/ort_config.json

@@ -0,0 +1,35 @@
+{
+  "one_external_file": true,
+  "opset": null,
+  "optimization": {},
+  "optimum_version": "1.17.0",
+  "quantization": {
+    "activations_dtype": "QUInt8",
+    "activations_symmetric": false,
+    "format": "QOperator",
+    "is_static": false,
+    "mode": "IntegerOps",
+    "nodes_to_exclude": [],
+    "nodes_to_quantize": [],
+    "operators_to_quantize": [
+      "Conv",
+      "MatMul",
+      "Attention",
+      "LSTM",
+      "Gather",
+      "Transpose",
+      "EmbedLayerNormalization"
+    ],
+    "per_channel": false,
+    "qdq_add_pair_to_weight": false,
+    "qdq_dedicated_pair": false,
+    "qdq_op_type_per_channel_support_to_axis": {
+      "MatMul": 1
+    },
+    "reduce_range": false,
+    "weights_dtype": "QUInt8",
+    "weights_symmetric": true
+  },
+  "transformers_version": "4.38.2",
+  "use_external_data_format": false
+}

model_fr_onnx/preprocessor_config.json

@@ -0,0 +1,10 @@
+{
+  "do_normalize": true,
+  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "processor_class": "Wav2Vec2Processor",
+  "return_attention_mask": false,
+  "sampling_rate": 16000
+}

model_fr_onnx/special_tokens_map.json

@@ -0,0 +1,28 @@
+{
+  "additional_special_tokens": [
+    {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    }
+  ],
+  "bos_token": "<s>",
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "[PAD]",
+  "unk_token": "[UNK]"
+}

model_fr_onnx/tokenizer_config.json

@@ -0,0 +1,52 @@
+{
+  "added_tokens_decoder": {
+    "57": {
+      "content": "[UNK]",
+      "lstrip": true,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": false
+    },
+    "58": {
+      "content": "[PAD]",
+      "lstrip": true,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": false
+    },
+    "59": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "60": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<s>",
+    "</s>"
+  ],
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": true,
+  "do_lower_case": false,
+  "eos_token": "</s>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "[PAD]",
+  "processor_class": "Wav2Vec2Processor",
+  "replace_word_delimiter_char": " ",
+  "target_lang": null,
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "unk_token": "[UNK]",
+  "word_delimiter_token": "|"
+}

model_fr_onnx/vocab.json

@@ -0,0 +1,61 @@
+{
+  "1": 1,
+  "[PAD]": 58,
+  "[UNK]": 57,
+  "a": 2,
+  "b": 3,
+  "d": 4,
+  "e": 5,
+  "f": 6,
+  "h": 7,
+  "i": 8,
+  "j": 9,
+  "k": 10,
+  "l": 11,
+  "m": 12,
+  "n": 13,
+  "o": 14,
+  "p": 15,
+  "r": 16,
+  "s": 17,
+  "t": 18,
+  "u": 19,
+  "v": 20,
+  "w": 21,
+  "x": 22,
+  "y": 23,
+  "z": 24,
+  "|": 0,
+  "ç": 25,
+  "ð": 26,
+  "ø": 27,
+  "ŋ": 28,
+  "œ": 29,
+  "ɐ": 30,
+  "ɑ": 31,
+  "ɒ": 32,
+  "ɔ": 33,
+  "ə": 34,
+  "ɛ": 35,
+  "ɜ": 36,
+  "ɡ": 37,
+  "ɣ": 38,
+  "ɨ": 39,
+  "ɪ": 40,
+  "ɬ": 41,
+  "ɲ": 42,
+  "ɹ": 43,
+  "ɾ": 44,
+  "ʁ": 45,
+  "ʃ": 46,
+  "ʊ": 47,
+  "ʌ": 48,
+  "ʍ": 49,
+  "ʒ": 50,
+  "ʔ": 51,
+  "ʲ": 52,
+  "ː": 53,
+  "̃": 54,
+  "β": 55,
+  "θ": 56
+}

requirements.txt

@@ -0,0 +1,9 @@
+fastapi
+python-multipart
+uvicorn
+torch
+numpy
+librosa
+optimum
+onnxruntime
+transformers