--- license: apache-2.0 library_name: litert pipeline_tag: text-classification tags: - mediapipe - litert - tflite - text-classification - on-device - android - ios - web - python - flutter - nlp - language-identification - multilingual - fp32 --- # MediaPipe Language Detector Ultra-lightweight UTF-8 text language identification model (`language_detector.tflite`, 315 KB, FP32) recognizing **110 languages** via ISO 639-1 / BCP-47 script codes (`af` through `zu`, plus `unknown`) in the [MediaPipe Language Detector (`LanguageDetector`)](https://developers.google.com/edge/mediapipe/solutions/text/language_detector) task.

Documentation: MediaPipe Language Detector Docs

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

Try live: Web Demo

Embedding-based neural classification network optimized with feature hashing and compact embedding lookup tables. Takes a single UTF-8 string (`Placeholder` `[1]`) and outputs a 111-way probability distribution (`pod/Softmax` `[1, 111]`), bundled with [`labels.txt`](./labels.txt) (`unknown` + 110 ISO 639-1 / `-Latn` transliterated language codes). ## Files | File | Precision | Size | SHA-256 | |---|---|---:|---| | [`language_detector.tflite`](https://huggingface.co/litert-community/MediaPipe-LanguageDetector/blob/main/language_detector.tflite) | `float32` | 315.3 KB (315,294 B) | `7db4f23dfe1ad8966b050b419a865da451143fd43eb6b606a256aadeeb1e5417` | ### Companion Files in Repo Root | File | Role | Size | SHA-256 | |---|---|---:|---| | [`LanguageDetector Model Card.pdf`](./LanguageDetector%20Model%20Card.pdf) | Upstream Google model card (PDF) | 250.2 KB (250,203 B) | `b383b7d189279949e1cc27aa04d09d967989eacd784cf91188a19c8811b69152` | | [`labels.txt`](./labels.txt) | Label map | 0.4 KB (376 B) | `938a527f1ed52b8a10cfb5491ac3229e60baae5befbbc7d3a60863917d0a0841` | ### Download ```bash pip install -U huggingface_hub hf download litert-community/MediaPipe-LanguageDetector language_detector.tflite --local-dir . ``` ## Inputs, Outputs, and Model Specifications ### `language_detector.tflite` - Embedded `TFLITE_METADATA`: name `LangID` · version `1` · author `MediaPipe` · packed associated files: `labels.txt` | Direction | Tensor | Shape | Type | Notes | |---|---|---|---|---| | Input | `Placeholder` | `[1]` | `string` | Placeholder | | Output | `pod/Softmax` | `[1, 111]` | `float32` | score — Prediction scores for each language. — Labels: `labels.txt` | ## Use with MediaPipe Tasks ### Python Quickstart ```bash pip install -U mediapipe huggingface_hub ``` ```python import mediapipe as mp from huggingface_hub import hf_hub_download model_path = hf_hub_download(repo_id="litert-community/MediaPipe-LanguageDetector", filename="language_detector.tflite") BaseOptions = mp.tasks.BaseOptions LanguageDetector = mp.tasks.text.LanguageDetector LanguageDetectorOptions = mp.tasks.text.LanguageDetectorOptions options = LanguageDetectorOptions( base_options=BaseOptions(model_asset_path=model_path), score_threshold=0.5, ) with LanguageDetector.create_from_options(options) as detector: result = detector.detect("Bonjour tout le monde, comment allez-vous ?") for det in result.detections: print(f"{det.language_code}: {det.probability:.3f}") ``` ### Cross-Platform Guides & Samples - **Task overview & options:** [Language Detector documentation](https://developers.google.com/edge/mediapipe/solutions/text/language_detector) - **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/#/text/language_detector) · [Web Source](https://github.com/google-ai-edge/mediapipe-samples-web/blob/main/src/tasks/language-detector.ts) · [Android](https://github.com/google-ai-edge/mediapipe-samples/tree/main/examples/language_detector/android) · [iOS](https://github.com/google-ai-edge/mediapipe-samples/tree/main/examples/language_detector/ios) · [Python Notebook](https://colab.research.google.com/github/google-ai-edge/mediapipe-samples/blob/main/examples/language_detector/python/%5BMediaPipe_Python_Tasks%5D_Language_Detector.ipynb) · [Flutter](https://github.com/google/flutter-mediapipe/tree/main/samples/text/language_detector) ## Latency Benchmarks End-to-end MediaPipe Task pipeline average latency reported by the [Language Detector documentation](https://developers.google.com/edge/mediapipe/solutions/text/language_detector#benchmarks): | Model Variant | Device | CPU Latency | GPU Latency | |---|---|---:|---:| | Language Detector | Pixel 6 | 0.31 ms | — | Additionally, the bundled PDF model card reports **90 µs** average latency on Pixel 2 (350-token text) with **F1 = 0.9781** (Precision 0.9893, Recall 0.9673) on Creative Commons v2.5 `en.wikinews.org` multilingual samples. ## Intended Use, Limitations, and Fairness Summarized from the bundled [`LanguageDetector Model Card.pdf`](./LanguageDetector%20Model%20Card.pdf) (Google Research Learn2Compress & MediaPipe Teams, May 10, 2023): - **Intended applications:** Predicting the language of input text (110 supported languages) as preprocessing for Smart Reply, OCR, and downstream NLP pipelines. - **Training data & bias:** Trained on PII-stripped translated anonymized web data covering languages across all 17 UN geographic subregions; has more training data for popular languages such as English and French. - **Out-of-scope uses:** - Not appropriate for **human life-critical decisions**, including financial, contractual, legal, or medical decisions. - Predicting languages outside the 110 preset codes in `labels.txt`. - Mixed multilingual strings (no guarantee that all languages or even the dominant language will be ranked first). ## 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 | |---|---|---|---|---| | `language_detector.tflite` | [`language_detector.tflite`](https://storage.googleapis.com/mediapipe-models/language_detector/language_detector/float32/1/language_detector.tflite) | `1682624733160468` | Thu, 27 Apr 2023 19:45:33 GMT | — | ## License and Citation - License: `apache-2.0` — Stated as Apache License, Version 2.0 in the bundled upstream Google model-card PDF.