---
library_name: transformers
license: apache-2.0
language:
- en
base_model:
- HuggingFaceTB/SmolLM2-360M
pipeline_tag: text-to-speech
---
# YarnGPT2b
## Table of Contents
1. [Model Summary](#model-summary)
2. [Model Description](#model-description)
3. [Bias, Risks, and Limitations](#bias-risks-and-limitations)
- [Recommendations](#recommendations)
4. [Speech Samples](#speech-samples)
5. [Training](#training)
6. [Future Improvements](#future-improvements)
7. [Citation](#citation)
8. [Credits & References](#credits--references)
## Model Summary
YarnGPT2b is a text-to-speech (TTS) and automatic speech recognition (ASR) model designed to synthesize Nigerian-accented languages (Yoruba, Igbo, Hausa, and English). It leverages pure language modeling without external adapters or complex architectures, providing high-quality, natural, and culturally relevant speech synthesis.
The model was trained on both TTS and ASR to explore whether learning patterns from one task could improve the other. However, this approach did not yield significant improvements, especially in ASR. This may be due to the small model size or the fact that the base model (YarnGPT2) was already highly optimized for TTS, making it difficult to learn ASR effectively.
#### How to use (Colab)
The model can generate audio on its own but its better to use a voice to prompt the model:
##### Voices (arranged in order of perfomance and stability)
- English: idera, chinenye, jude, emma,umar,,joke,zainab ,osagie, remi, tayo
- Yoruba: yoruba_male2, yoruba_female2, yoruba_feamle1
- Igbo: igbo_female2, igbo_male2,igbo_female1,
- Hausa: hausa_feamle1,hausa_female2, hausa_male2,hausa_male1
### Prompt YarnGPT2b
```python
!git clone https://github.com/saheedniyi02/yarngpt.git
pip install outetts uroman
import os
import re
import json
import torch
import inflect
import random
import uroman as ur
import numpy as np
import torchaudio
import IPython
from transformers import AutoModelForCausalLM, AutoTokenizer
from outetts.wav_tokenizer.decoder import WavTokenizer
!wget https://huggingface.co/novateur/WavTokenizer-medium-speech-75token/resolve/main/wavtokenizer_mediumdata_frame75_3s_nq1_code4096_dim512_kmeans200_attn.yaml
!gdown 1-ASeEkrn4HY49yZWHTASgfGFNXdVnLTt
from yarngpt.audiotokenizer import AudioTokenizerV2
tokenizer_path="saheedniyi/YarnGPT2b"
wav_tokenizer_config_path="/content/wavtokenizer_mediumdata_frame75_3s_nq1_code4096_dim512_kmeans200_attn.yaml"
wav_tokenizer_model_path = "/content/wavtokenizer_large_speech_320_24k.ckpt"
audio_tokenizer=AudioTokenizerV2(
tokenizer_path,wav_tokenizer_model_path,wav_tokenizer_config_path
)
model = AutoModelForCausalLM.from_pretrained(tokenizer_path,torch_dtype="auto").to(audio_tokenizer.device)
#change the text
text="The election was won by businessman and politician, Moshood Abiola, but Babangida annulled the results, citing concerns over national security."
# change the language and voice
prompt=audio_tokenizer.create_prompt(text,lang="english",speaker_name="idera")
input_ids=audio_tokenizer.tokenize_prompt(prompt)
output = model.generate(
input_ids=input_ids,
temperature=0.1,
repetition_penalty=1.1,
max_length=4000,
#num_beams=5,# using a beam size helps for the local languages but not english
)
codes=audio_tokenizer.get_codes(output)
audio=audio_tokenizer.get_audio(codes)
IPython.display.Audio(audio,rate=24000)
torchaudio.save(f"Sample.wav", audio, sample_rate=24000)
```
### Simple Nigerian Accented-NewsReader
```python
!git clone https://github.com/saheedniyi02/yarngpt.git
pip install outetts uroman trafilatura pydub
import os
import re
import json
import torch
import inflect
import random
import requests
import trafilatura
import inflect
import uroman as ur
import numpy as np
import torchaudio
import IPython
from pydub import AudioSegment
from pydub.effects import normalize
from transformers import AutoModelForCausalLM, AutoTokenizer
from outetts.wav_tokenizer.decoder import WavTokenizer
!wget https://huggingface.co/novateur/WavTokenizer-medium-speech-75token/resolve/main/wavtokenizer_mediumdata_frame75_3s_nq1_code4096_dim512_kmeans200_attn.yaml
!gdown 1-ASeEkrn4HY49yZWHTASgfGFNXdVnLTt
from yarngpt.audiotokenizer import AudioTokenizerV2
tokenizer_path="saheedniyi/YarnGPT2b"
wav_tokenizer_config_path="/content/wavtokenizer_mediumdata_frame75_3s_nq1_code4096_dim512_kmeans200_attn.yaml"
wav_tokenizer_model_path = "/content/wavtokenizer_large_speech_320_24k.ckpt"
audio_tokenizer=AudioTokenizerV2(
tokenizer_path,wav_tokenizer_model_path,wav_tokenizer_config_path
)
model = AutoModelForCausalLM.from_pretrained(tokenizer_path,torch_dtype="auto").to(audio_tokenizer.device)
# Split text into chunks
def split_text_into_chunks(text, word_limit=25):
sentences=[sentence.strip() for sentence in text.split('.') if sentence.strip()]
chunks=[]
for sentence in sentences:
chunks.append(".")
sentence_splitted=sentence.split(" ")
num_words=len(sentence_splitted)
if (num_words>word_limit) and (num_words<=word_limit*2):
chunks.append(" ".join(sentence_splitted[:int(num_words/2)]))
chunks.append(" ".join(sentence_splitted[int(num_words/2):]))
elif (num_words>word_limit*2) and (num_words<=word_limit*3):
chunks.append(" ".join(sentence_splitted[:int(num_words/3)]))
chunks.append(" ".join(sentence_splitted[int(num_words/3):int(2*num_words/3)]))
chunks.append(" ".join(sentence_splitted[int(2*num_words/3):]))
elif (num_words>word_limit*3) and (num_words<=word_limit*4):
chunks.append(" ".join(sentence_splitted[:int(num_words/4)]))
chunks.append(" ".join(sentence_splitted[int(num_words/4):word_limit*2]))
chunks.append(" ".join(sentence_splitted[int(2*num_words/4):int(3*num_words/4)]))
chunks.append(" ".join(sentence_splitted[int(3*num_words/4):]))
elif (num_words>word_limit*4) and (num_words<=word_limit*5):
chunks.append(" ".join(sentence_splitted[:int(num_words/5)]))
chunks.append(" ".join(sentence_splitted[int(num_words/5):int(2*num_words/5)]))
chunks.append(" ".join(sentence_splitted[int(2*num_words/5):int(3*num_words/5)]))
chunks.append(" ".join(sentence_splitted[int(3*num_words/5):int(4*num_words/5)]))
chunks.append(" ".join(sentence_splitted[int(4*num_words/5):]))
else:
chunks.append(sentence)
return chunks
def speed_change(sound, speed=0.9):
# Manually override the frame_rate. This tells the computer how many
# samples to play per second
sound_with_altered_frame_rate = sound._spawn(sound.raw_data, overrides={
"frame_rate": int(sound.frame_rate * speed)
})
# convert the sound with altered frame rate to a standard frame rate
# so that regular playback programs will work right. They often only
# know how to play audio at standard frame rate (like 44.1k)
return sound_with_altered_frame_rate.set_frame_rate(sound.frame_rate)
#change the url
url="https://punchng.com/im-not-desperate-for-2027-presidential-ticket-obi/"
page=requests.get(url)
content=trafilatura.extract(page.text)
chunks=split_text_into_chunks(content)
all_codes=[]
#Looping over the chunks and adding creating a large `all_codes` list
for i,chunk in enumerate(chunks):
print(i)
print("\n")
print(chunk)
if chunk==".":
#add silence for 0.5 seconds if we encounter a full stop
all_codes.extend([453]*38)
else:
# Change the language and voice here
prompt=audio_tokenizer.create_prompt(chunk,lang="english",speaker_name="jude")
input_ids=audio_tokenizer.tokenize_prompt(prompt)
output = model.generate(
input_ids=input_ids,
temperature=0.1,
repetition_penalty=1.1,
max_length=4000,
#num_beams=5,
)
codes=audio_tokenizer.get_codes(output)
all_codes.extend(codes)
audio=audio_tokenizer.get_audio(all_codes)
IPython.display.Audio(audio,rate=24000)
torchaudio.save(f"news1.wav",
audio,
sample_rate=24000,
)
```
## Model Description
- **Developed by:** [Saheedniyi](https://linkedin.com/in/azeez-saheed)
- **Model type:** Text-to-Speech
- **Language(s) (NLP):** English--> Nigerian Accented English
- **Finetuned from:** [HuggingFaceTB/SmolLM2-360M](https://huggingface.co/HuggingFaceTB/SmolLM2-360M)
- **Repository:** [YarnGPT Github Repository](https://github.com/saheedniyi02/yarngpt)
- **Paper:** IN PROGRESS.
- **Demo:** 1) [Prompt YarnGPT2b notebook](https://colab.research.google.com/drive/13-o1X5F3CLeHixjqobNf2TJN1T6LWOqx?usp=sharing)
2) [Simple news reader](https://colab.research.google.com/drive/1FLTUmESJbG52Bj21XX3-AoevjaXwtmhE?usp=sharing)
#### Uses
Generate Nigerian-accented English speech for experimental purposes.
#### Out-of-Scope Use
The model is not suitable for generating speech in languages other than English or other accents.
## Bias, Risks, and Limitations
The model may not capture the full diversity of Nigerian accents and could exhibit biases based on the training dataset. Also a lot of the text the model was trained on were automatically generated which could impact performance.
#### Recommendations
Users (both direct and downstream) should be made aware of the risks, biases, and limitations of the model. Feedback and diverse training data contributions are encouraged.
## Speech Samples
Listen to samples generated by YarnGPT:
Input
Audio
Notes
Uhm, so, what was the inspiration behind your latest project? Like, was there a specific moment where you were like, 'Yeah, this is it!' Or, you know, did it just kind of, uh, come together naturally over time
Habeeb Okikiọla Olalomi Badmus ti ọpọ awọn ololufẹ rẹ mọ si Portable ti sọ fun ile ẹjọ majisireeti ti ipinlẹ Ogun wi pe ṣaka lara oun da, oun ko ni aisan tabi arun kankan lara.
Isi ụlọorụ Shell dị na Lọndọn na gọọmenti Naịjirịa ekwuputala ugboro ugboro na ọrụ ịsacha ogbe ndị lara n'iyi n'Ogoni bụ nke malitere ihe dịka afọ asatọ gara aga na-aga nke ọma.
Bisa ga dukkan alamu, haƙata cimma ruwa, dangane da koke-koken da tsofaffin ma'aikatan tarayya ke ta yi, a kan dimbin basukan wasu hakkokinsu da suke bi shekara da shekaru.
## Training
#### Data
Trained on a dataset of publicly available Nigerian movies, podcasts ( using the subtitle-audio pairs) and open source Nigerian-related audio data on Huggingface,
#### Preprocessing
Audio files were preprocessed and resampled to 24Khz and tokenized using [wavtokenizer](https://huggingface.co/novateur/WavTokenizer).
#### Training Hyperparameters
- **Number of epochs:** 5
- **batch_size:** 4
- **Scheduler:** linear schedule with warmup for 4 epochs, then linear decay to zero for the last epoch
- **Optimizer:** AdamW (betas=(0.9, 0.95),weight_decay=0.01)
- **Learning rate:** 1*10^-3
#### Hardware
- **GPUs:** 1 A100 (google colab: 50 hours)
#### Software
- **Training Framework:** Pytorch
## Future Improvements?
- Scaling up model size and human-annotaed/ reviewed training data
- Wrap the model around an API endpoint
- Add support for local Nigerian languages
- Voice cloning.
- Potential expansion into speech-to-speech assistant models
## Citation [optional]
#### BibTeX:
```python
@misc{yarngpt2025,
author = {Saheed Azeez},
title = {YarnGPT: Nigerian-Accented English Text-to-Speech Model},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/SaheedAzeez/yarngpt}
}
```
#### APA:
```python
Saheed Azeez. (2025). YarnGPT: Nigerian-Accented English Text-to-Speech Model. Hugging Face. Available at: https://huggingface.co/saheedniyi/YarnGPT
```
## Credits & References
- [OuteAI/OuteTTS-0.2-500M](https://huggingface.co/OuteAI/OuteTTS-0.2-500M/)
- [WavTokenizer](https://github.com/jishengpeng/WavTokenizer)
- [CTC Forced Alignment](https://pytorch.org/audio/stable/tutorials/ctc_forced_alignment_api_tutorial.html)
- [Voicera](https://huggingface.co/Lwasinam/voicera)