Upload PretrainedAudioMAEEncoder

README.md

@@ -0,0 +1,199 @@
+---
+library_name: transformers
+tags: []
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]

config.json

@@ -0,0 +1,18 @@
+{
+  "architectures": [
+    "PretrainedAudioMAEEncoder"
+  ],
+  "auto_map": {
+    "AutoConfig": "model.AudioMAEConfig",
+    "AutoModel": "model.PretrainedAudioMAEEncoder"
+  },
+  "dtype": "float32",
+  "img_size": [
+    1024,
+    128
+  ],
+  "in_chans": 1,
+  "model_type": "audiomae",
+  "num_classes": 0,
+  "transformers_version": "4.57.3"
+}

model.py

@@ -0,0 +1,147 @@
+from typing import Tuple
+
+import torch
+import torchaudio
+import torchaudio.transforms as transforms
+from torchaudio.compliance import kaldi
+from transformers import PretrainedConfig
+
+from einops import rearrange
+
+from timm.models.vision_transformer import VisionTransformer
+from transformers import PreTrainedModel
+
+
+# it seems like Config class and Model class should be located in the same file; otherwise, seemingly casuing an issue in model loading after pushing to HF.
+class AudioMAEConfig(PretrainedConfig):
+    model_type = "audiomae"
+
+    def __init__(self,
+                 img_size:Tuple[int,int]=(1024,128),
+                 in_chans:int=1,
+                 num_classes:int=0,
+                 **kwargs,):
+        super().__init__(**kwargs)
+        self.img_size = img_size
+        self.in_chans = in_chans
+        self.num_classes = num_classes
+
+
+class AudioMAEEncoder(VisionTransformer):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        """
+        - img_size of (1024, 128) = (temporal_length, n_freq_bins) is fixed, as described in the paper
+        - AudoMAE accepts a mono-channel (i.e., in_chans=1)
+        """
+        self.MEAN = -4.2677393  # written on the paper
+        self.STD = 4.5689974  # written on the paper
+
+    def load_wav_file(self, file_path:str):
+        """
+        to use this, `torchaudio` and `ffmpeg` must be installed
+        - `ffmpeg` version must be >=4.4 and <7.
+        - `ffmpeg` installation by `conda install -c conda-forge ffmpeg==6.1.1`
+        """
+        audio, sample_rate = torchaudio.load(file_path)  # audio: (n_channels, length); 
+
+
+        # Check if the audio has multiple channels
+        if audio.shape[0] > 1:
+            # Convert stereo audio to mono by taking the mean across channels
+            # AudioMAE accepts a mono channel.
+            audio = torch.mean(audio, dim=0, keepdim=True)
+
+        # resample the audio into 16khz
+        # AudioMAE accepts 16khz
+        if sample_rate != 16000:
+            converter = transforms.Resample(orig_freq=sample_rate, new_freq=16000)
+            audio = converter(audio)
+
+        # length clip
+        audio_len = audio.shape[-1] / 16000
+        if audio_len > 10.0:
+            print(f'{file_path} has audio length {audio_len}s, which is longer than 10s. It will be segmented into 10-second windows with a 5-second stride (50% overlap)')
+            # current sampling rate is 16000, so 10 seconds is 160000 samples and 5 seconds is 80000 samples
+            window_size = 160000
+            stride = 80000
+            remainder = (audio.shape[-1] - window_size) % stride
+            if remainder != 0:
+                padding = (0, stride - remainder)
+                audio = torch.nn.functional.pad(audio, padding, "constant", 0)
+            audio = audio.squeeze(0).unfold(0, window_size, stride)
+            return audio
+        else:
+            return audio
+    
+    def waveform_to_melspec(self, waveform:torch.FloatTensor):
+        # Compute the Mel spectrogram using Kaldi-compatible features
+        # the parameters are chosen as described in the audioMAE paper (4.2 implementation details)
+        mel_spectrogram = kaldi.fbank(
+            waveform, 
+            num_mel_bins=128, 
+            frame_length=25.0, 
+            frame_shift=10.0, 
+            htk_compat=True, 
+            use_energy=False,
+            sample_frequency=16000, 
+            window_type='hanning',
+            dither=0.0
+        )
+        
+        # Ensure the output shape matches 1x1024x128 by padding or trimming the time dimension
+        expected_frames = 1024  # as described in the paper
+        current_frames = mel_spectrogram.shape[0]
+        if current_frames > expected_frames:
+            mel_spectrogram = mel_spectrogram[:expected_frames, :]
+        elif current_frames < expected_frames:
+            padding = expected_frames - current_frames
+            mel_spectrogram = torch.nn.functional.pad(mel_spectrogram, (0, 0,  # (left, right) for the 1st dim
+                                                                        0, padding),  # (left, right) for the 2nd dim
+                                                                        )
+            
+        # scale
+        # as in the AudioMAE implementation [REF: https://github.com/facebookresearch/AudioMAE/blob/bd60e29651285f80d32a6405082835ad26e6f19f/dataset.py#L300]
+        mel_spectrogram = (mel_spectrogram - self.MEAN) / (self.STD * 2)  # (length, n_freq_bins) = (1024, 128)
+        return mel_spectrogram
+
+    @torch.no_grad()
+    def encode(self, file_path:str, device, reshape=True):
+        self.eval()
+
+        waveform = self.load_wav_file(file_path)
+
+        zs = []
+        for i in range(waveform.shape[0]):
+            melspec = self.waveform_to_melspec(waveform[i].unsqueeze(0))  # (length, n_freq_bins) = (1024, 128)
+            melspec = melspec[None,None,:,:]  # (1, 1, length, n_freq_bins) = (1, 1, 1024, 128)
+            z = self.forward_features(melspec.to(device)).cpu()  # (b, 1+n, d); d=768
+            z = z[:,1:,:]  # (b n d); remove [CLS], the class token
+            b, c, w, h = melspec.shape  # w: temporal dim; h:freq dim
+            if reshape:
+                wprime = round(w / self.patch_embed.patch_size[0])  # width in the latent space
+                hprime = round(h / self.patch_embed.patch_size[1])  # height in the latent space
+                # reconstruct the temporal and freq dims
+                z = rearrange(z, 'b (w h) d -> b d h w', h=hprime)  # (b d h' w')
+            else:
+                # put the patch dim in the back, i.e.,  (b n d)-> (b d n)
+                z = z.transpose(1, 2)
+
+            # remove the batch dim
+            z = z[0]  # (d h' w') if reshape else (d n)
+            zs.append(z)
+        z = torch.stack(zs, dim=0)
+        z = z.mean(dim=0).unsqueeze(0)
+        return z  # (d h' w') if reshape
+
+
+class PretrainedAudioMAEEncoder(PreTrainedModel):
+    config_class = AudioMAEConfig
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.encoder = AudioMAEEncoder(img_size=config.img_size, in_chans=config.in_chans, num_classes=config.num_classes)
+    
+    def forward(self, file_path:str, reshape=True):
+        device = self.device
+        return self.encoder.encode(file_path, device, reshape)  # (d h' w')

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c868088d7f6f9ee8c29292bfa029a552bad781e1c59abb3692762330811bf535
+size 342607672