Upload distilled conformer recognizer object and loader script

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+recognizer.dill filter=lfs diff=lfs merge=lfs -text

modeling.py

@@ -0,0 +1,62 @@
+# Auto-generated to contain necessary class definitions for loading the recognizer.
+import torch
+import torch.nn as nn
+import torchaudio
+import torchaudio.transforms as T
+from torchaudio.models import Conformer
+from allosaurus.audio import Audio
+
+class ConformerAcousticModel(nn.Module):
+    def __init__(self, input_dim: int, num_phonemes: int, d_model: int, ffn_dim: int = 2560, num_heads: int = 4, num_layers: int = 8, depthwise_conv_kernel_size: int = 31, dropout: float = 0.1):
+        super().__init__()
+        self.input_projection = nn.Sequential(
+            nn.Linear(input_dim, d_model),
+            nn.LayerNorm(d_model),
+            nn.Dropout(dropout)
+        )
+        self.conformer = Conformer(
+            input_dim=d_model,
+            num_heads=num_heads,
+            ffn_dim=ffn_dim,
+            num_layers=num_layers,
+            depthwise_conv_kernel_size=depthwise_conv_kernel_size,
+            dropout=dropout
+        )
+        self.output_projection = nn.Linear(d_model, num_phonemes)
+
+    def forward(self, features: torch.Tensor, lengths: torch.Tensor) -> torch.Tensor:
+        x = self.input_projection(features)
+        x, _ = self.conformer(x, lengths)
+        logits = self.output_projection(x)
+        return logits
+
+class UpgradedRecognizer:
+    def __init__(self, pm_module, am_module, lm_module, device):
+        self.pm = pm_module
+        self.am = am_module
+        self.lm = lm_module
+        self.device = device
+        self.am.to(self.device)
+        self.am.eval()
+
+    def recognize(self, audio_path: str) -> str:
+        waveform, sr = torchaudio.load(audio_path)
+        if sr != 16000:
+            resampler = T.Resample(sr, 16000).to(waveform.device)
+            waveform = resampler(waveform)
+        if waveform.shape[0] > 1:
+            waveform = torch.mean(waveform, dim=0, keepdim=True)
+        
+        audio_object = Audio(waveform.squeeze().cpu().numpy(), 16000)
+        
+        features = self.pm.compute(audio_object)
+        features_tensor = torch.tensor(features).unsqueeze(0).to(self.device)
+        lengths_tensor = torch.tensor([features_tensor.shape[1]], device=self.device)
+        
+        with torch.no_grad():
+            logits = self.am(features_tensor, lengths_tensor)
+
+        logits_numpy = logits.squeeze(0).cpu().numpy()
+        phoneme_list = self.lm.compute(logits_numpy, lang_id='ipa', topk=1)
+        
+        return " ".join(phoneme_list)

recognizer.dill

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b7fe6beee958af085db2ceb6fe6c30b1666a48ca8b710ef6bcdbea2e20faf5a
+size 190360911