--- license: apache-2.0 library_name: litert pipeline_tag: audio-classification tags: - mediapipe - litert - tflite - audio-classification - on-device - android - ios - web - python - flutter - audio - audioset - yamnet - sound-event-detection - fp32 - int8 --- # MediaPipe YAMNet (Audio Classifier) Audio event classification model (`yamnet.tflite`, 4.1 MB, float32 PCM I/O with internal 8-bit weight quantization) trained on AudioSet to recognize **521 sound classes** (speech, musical instruments, animals, alarms, environmental sounds) in the [MediaPipe Audio Classifier (`AudioClassifier`)](https://developers.google.com/edge/mediapipe/solutions/audio/audio_classifier) task.

Documentation: MediaPipe Audio Classifier Docs

Samples: Android · iOS · Python Colab · Web · Flutter

Try live: Web Demo

Takes 16 kHz mono float32 waveform windows (`15600` samples = `0.975 s`, normalized in `[-1.0, 1.0]`), computes log-mel spectrograms internally, and outputs `[1, 521]` class scores aligned with [`yamnet_label_list.txt`](./yamnet_label_list.txt) (embedded inside `TFLITE_METADATA` and included as a companion file). ## Files | File | Precision | Size | SHA-256 | |---|---|---:|---| | [`yamnet.tflite`](https://huggingface.co/litert-community/MediaPipe-AudioClassifier-YAMNet/blob/main/yamnet.tflite) | `float32` | 4.13 MB (4,126,810 B) | `4d8b4a53282dc83ef04e3e7dbc4fbc98082e34e44ed798e16c3a0cdd4c584faf` | ### Companion Files in Repo Root | File | Role | Size | SHA-256 | |---|---|---:|---| | [`yamnet_label_list.txt`](./yamnet_label_list.txt) | Label map | 6.2 KB (6,230 B) | `8e1267a120c1932b7273c0d0e0c5529edbb9a35512b437b1c8982baa59047051` | ### Download ```bash pip install -U huggingface_hub hf download litert-community/MediaPipe-AudioClassifier-YAMNet yamnet.tflite --local-dir . ``` ## Inputs, Outputs, and Model Specifications ### `yamnet.tflite` - Embedded `TFLITE_METADATA`: name `yamnet/classification` · version `1` · author `MediaPipe` · packed associated files: `yamnet_label_list.txt` | Direction | Tensor | Shape | Type | Notes | |---|---|---|---|---| | Input | `waveform_binary` | `[15600]` | `float32` | audio_clip — Input audio clip to be classified (16 kHz float32 waveform samples in range -1.0..1.0). | | Output | `tower0/network/layer32/final_output` | `[1, 521]` | `float32` | scores — Scores in range 0..1.0 for each of the 521 output classes. — Labels: `yamnet_label_list.txt` | ## Use with MediaPipe Tasks ### Python Quickstart ```bash pip install -U mediapipe huggingface_hub ``` ```python import numpy as np import mediapipe as mp from huggingface_hub import hf_hub_download model_path = hf_hub_download(repo_id="litert-community/MediaPipe-AudioClassifier-YAMNet", filename="yamnet.tflite") BaseOptions = mp.tasks.BaseOptions AudioClassifier = mp.tasks.audio.AudioClassifier AudioClassifierOptions = mp.tasks.audio.AudioClassifierOptions AudioRunningMode = mp.tasks.audio.RunningMode AudioData = mp.tasks.components.containers.AudioData options = AudioClassifierOptions( base_options=BaseOptions(model_asset_path=model_path), running_mode=AudioRunningMode.AUDIO_CLIPS, max_results=5, ) clip_samples = np.zeros(16000, dtype=np.float32) # replace with 16 kHz mono PCM in [-1.0, 1.0] audio_data = AudioData.create_from_array(clip_samples, sample_rate=16000) with AudioClassifier.create_from_options(options) as classifier: for window in classifier.classify(audio_data): print([(c.category_name, round(c.score, 3)) for c in window.classifications[0].categories]) ``` ### Cross-Platform Guides & Samples - **Task overview & options:** [Audio Classifier documentation](https://developers.google.com/edge/mediapipe/solutions/audio/audio_classifier) - **Platform setup guides:** [Python](https://developers.google.com/edge/mediapipe/solutions/setup_python) · [Web / JavaScript](https://developers.google.com/edge/mediapipe/solutions/setup_web) · [Android (Kotlin / Java)](https://developers.google.com/edge/mediapipe/solutions/setup_android) · [iOS (Swift)](https://developers.google.com/edge/mediapipe/solutions/setup_ios) · [Flutter (Dart)](https://developers.google.com/edge/mediapipe/solutions/setup_flutter) - **Runnable sample apps:** [Web Demo](https://google-ai-edge.github.io/mediapipe-samples-web/#/audio/audio_classifier) · [Web Source](https://github.com/google-ai-edge/mediapipe-samples-web/blob/main/src/tasks/audio-classifier.ts) · [Android](https://github.com/google-ai-edge/mediapipe-samples/tree/main/examples/audio_classifier/android) · [iOS](https://github.com/google-ai-edge/mediapipe-samples/tree/main/examples/audio_classifier/ios) · [Python Notebook](https://colab.research.google.com/github/google-ai-edge/mediapipe-samples/blob/main/examples/audio_classifier/python/audio_classification.ipynb) · [Flutter](https://github.com/google/flutter-mediapipe/tree/main/samples/audio/audio_classifier) ## Latency Benchmarks End-to-end MediaPipe Task pipeline average latency reported by the [Audio Classifier documentation](https://developers.google.com/edge/mediapipe/solutions/audio/audio_classifier#benchmarks): | Model Variant | Device | CPU Latency | GPU Latency | |---|---|---:|---:| | YamNet | Pixel 6 | 12.29 ms | — | ## Intended Use, Limitations, and Fairness - **Intended applications:** Real-time audio event tagging and acoustic scene recognition across the 521 AudioSet categories on mobile, web, desktop, and embedded devices. - **Limitations & considerations:** Multi-label sound event classifier (not a speech-to-text ASR model or biometric speaker recognizer); expects 16 kHz mono audio normalized to `[-1.0, 1.0]`. No standalone Google fairness PDF is published on the MediaPipe documentation page. ## Provenance Every model binary in this repository is byte-identical (unmodified FlatBuffer weights and embedded `TFLITE_METADATA` bytes) to the official artifact served by Google from `storage.googleapis.com/mediapipe-models`. | Repo File | Upstream URL | GCS Generation | Upstream `Last-Modified` | Pinned Bucket Version | |---|---|---|---|---| | `yamnet.tflite` | [`yamnet.tflite`](https://storage.googleapis.com/mediapipe-models/audio_classifier/yamnet/float32/latest/yamnet.tflite) | `1682545011390892` | Wed, 26 Apr 2023 21:36:51 GMT | `/1/` | ## License and Citation - License: `apache-2.0` — Stated as Apache License, Version 2.0 (`http://www.apache.org/licenses/LICENSE-2.0`) in the embedded `TFLITE_METADATA` block of the `.tflite` model file(s). ### Citation Trained on [AudioSet](https://research.google.com/audioset/) (`TFLITE_METADATA` author: MediaPipe, Apache License 2.0).