---
license: bsd-2-clause
language:
- en
task: grapheme-to-phoneme-conversion
size: ~130,000 word-pronunciation pairs
format: "text-phoneme pairs"
---

# CMUdict - Carnegie Mellon Pronouncing Dictionary

This dataset contains grapheme-to-phoneme (G2P) mappings derived from the CMU Pronouncing Dictionary version 0.7b.

## Dataset Description

- **License** BSD 2-Clause License (CMU license)
- **Task** Grapheme-to-Phoneme Conversion
- **Language** English
- **Size** ~130,000 word-pronunciation pairs
- **Format** Text-phoneme pairs

## Usage

This dataset is structured for training and evaluating grapheme-to-phoneme (G2P) conversion models, which convert written words to their phonetic pronunciations.

### Loading the Dataset

```python
from datasets import load_dataset

# Load from Huggingface Hub
dataset = load_dataset("jc4p/CMUdict")

# Access splits
train_data = dataset["train"]
validation_data = dataset["validation"]
test_data = dataset["test"]

# Example
sample = train_data[0]
print(f"Word: {sample['text']}")
print(f"Phonemes: {sample['phonemes']}")
```

### Use Cases

1. **Text-to-Speech (TTS) Systems**
   - Train models to predict pronunciation for out-of-vocabulary words
   - Improve naturalness of synthetic speech by providing accurate pronunciations

2. **Automatic Speech Recognition (ASR)**
   - Create pronunciation dictionaries for language models
   - Improve recognition accuracy for rare or unusual words

3. **Language Learning Applications**
   - Help language learners with correct pronunciation
   - Create interactive pronunciation guides

4. **Linguistic Research**
   - Study phonological patterns in English
   - Compare dialectal variations in pronunciation

5. **Spelling Correction**
   - Use phonetic similarity to detect and correct spelling errors

### Training a G2P Model

Here's a basic example of how to train a sequence-to-sequence model:

```python
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, Seq2SeqTrainingArguments, Seq2SeqTrainer
from datasets import load_dataset

# Load dataset
dataset = load_dataset("CMUdict")

# Prepare tokenizer
tokenizer = AutoTokenizer.from_pretrained("t5-small")

# Tokenize function
def tokenize_function(examples):
    model_inputs = tokenizer(examples["text"], max_length=128, truncation=True)
    labels = tokenizer(examples["phonemes"], max_length=128, truncation=True)
    model_inputs["labels"] = labels["input_ids"]
    return model_inputs

# Apply tokenization
tokenized_datasets = dataset.map(tokenize_function, batched=True)

# Load model
model = AutoModelForSeq2SeqLM.from_pretrained("t5-small")

# Training arguments
training_args = Seq2SeqTrainingArguments(
    output_dir="./results",
    evaluation_strategy="epoch",
    learning_rate=5e-5,
    per_device_train_batch_size=16,
    per_device_eval_batch_size=16,
    weight_decay=0.01,
    save_total_limit=3,
    num_train_epochs=3,
    predict_with_generate=True,
)

# Initialize trainer
trainer = Seq2SeqTrainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets["train"],
    eval_dataset=tokenized_datasets["validation"],
    tokenizer=tokenizer,
)

# Train the model
trainer.train()
```

## License Information

```
Copyright (C) 1993-2015 Carnegie Mellon University. All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:

1. Redistributions of source code must retain the above copyright
   notice, this list of conditions and the following disclaimer.
   The contents of this file are deemed to be source code.

2. Redistributions in binary form must reproduce the above copyright
   notice, this list of conditions and the following disclaimer in
   the documentation and/or other materials provided with the
   distribution.

This work was supported in part by funding from the Defense Advanced
Research Projects Agency, the Office of Naval Research and the National
Science Foundation of the United States of America, and by member
companies of the Carnegie Mellon Sphinx Speech Consortium. We acknowledge
the contributions of many volunteers to the expansion and improvement of
this dictionary.

THIS SOFTWARE IS PROVIDED BY CARNEGIE MELLON UNIVERSITY ``AS IS'' AND
ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY
NOR ITS EMPLOYEES BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
```

## Citation

```
@misc{cmu_pronouncing_dictionary,
  title={The {CMU} Pronouncing Dictionary},
  author={{Carnegie Mellon University}},
  howpublished={\url{http://www.speech.cs.cmu.edu/cgi-bin/cmudict}},
  year={2015},
  note={Version 0.7b}
}
```