wav2vec2-xlsr-300m-german-truecase Open-source Model - Freely Implement German Speech Recognition While Preserving Capitalization

Wav2vec2 Xlsr 300m German Truecase

Developed by abnerh

Based on Facebook's wav2vec2-xls-r-300m model, fine-tuned on the Common Voice German dataset, supporting German speech recognition with preserved text case information.

Speech Recognition

Transformers

#German Speech Recognition #True Case Output #16kHz Sampling Rate

Downloads 16

Release Time : 3/2/2022

Model Overview

This model is a German automatic speech recognition (ASR) system that specifically handles case sensitivity in German, directly outputting text with proper German grammar case without post-processing.

Model Features

True Case Handling

The model directly learns German case rules, outputting text with proper grammar case without additional post-processing.

Fine-tuned on Common Voice

Fine-tuned using a high-quality open-source German speech dataset to improve German recognition accuracy.

16kHz Sampling Rate Support

Optimized for processing 16kHz sampled speech input, suitable for common speech application scenarios.

Model Capabilities

German Speech-to-Text

Automatic Case Conversion

Continuous Speech Recognition

Use Cases

Speech Transcription

Automated Meeting Minutes

Automatically convert German meeting recordings into text transcripts with correct case.

Reduce manual transcription workload while maintaining professional document formatting.

Subtitle Generation

Generate subtitles with correct case for German video content.

Enhance subtitle professionalism and readability.

Voice Assistants

German Voice Command Recognition

Recognize German voice commands in smart home or customer service systems.

Accurately understand German commands with case variations.

🚀 Fine-tuned Wav2Vec2-XLS-R-300M for German

This project fine-tunes the facebook/wav2vec2-xls-r-300m model on German using the Common Voice dataset. When using this model, ensure that your speech input is sampled at 16kHz.

Since capitalization is crucial in the German language (e.g., Sie vs. sie), a model was trained with a vocabulary that includes both lower and upper case letters. The hope was that the model would learn the correct casing, eliminating the need for post-processing like truecasing.

🚀 Quick Start

Model Usage

from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
import soundfile as sf
import torch
 
# load model and processor
processor = Wav2Vec2Processor.from_pretrained("abnerh/wav2vec2-xlsr-300m-german-truecase")
model = Wav2Vec2ForCTC.from_pretrained("abnerh/wav2vec2-xlsr-300m-german-truecase")
 
speech, sr = sf.read('audio.wav') 
# tokenize
input_values = processor(speech, return_tensors="pt", padding="longest").input_values  # Batch size 1

# retrieve logits
logits = model(input_values).logits
 
# take argmax and decode
predicted_ids = torch.argmax(logits, dim=-1)
transcription = processor.batch_decode(predicted_ids)

# print transcription results
print(transcription)

💻 Usage Examples

Basic Usage

from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
import soundfile as sf
import torch
 
# load model and processor
processor = Wav2Vec2Processor.from_pretrained("abnerh/wav2vec2-xlsr-300m-german-truecase")
model = Wav2Vec2ForCTC.from_pretrained("abnerh/wav2vec2-xlsr-300m-german-truecase")
 
speech, sr = sf.read('audio.wav') 
# tokenize
input_values = processor(speech, return_tensors="pt", padding="longest").input_values  # Batch size 1

# retrieve logits
logits = model(input_values).logits
 
# take argmax and decode
predicted_ids = torch.argmax(logits, dim=-1)
transcription = processor.batch_decode(predicted_ids)

# print transcription results
print(transcription)

📚 Documentation

Prediction Results

Reference	Prediction
Die zoologische Einordnung der Spezies ist seit Jahrzehnten umstritten	Die psoologische Einordnung der Spezies ist seit Jahrzehnten umstritten
Hauptgeschäftsfeld war ursprünglich der öffentliche Sektor in Irland	Hauptgeschäftsfeld war ursprünglich der öffentliche Sektor in Irland
Er vertrat den Wahlkreis Donauwörth im Parlament	Er vertrat den Wahlkreis DonauWört im Parlament
Ich bin gespannt welche Lieder sie wählt	Ich bin gespannt welche Lieder see wählt
Eine allgemein verbindliche Definition gibt es nicht	Eine allgemeinverbindliche Definition gibt es nicht

⚠️ Important Note

When using this model, make sure that your speech input is sampled at 16kHz.

💡 Usage Tip

Training the model with a vocabulary including both lower and upper case letters helps the model learn the correct casing, removing the need for post-processing like truecasing.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご