tiny-audio, nanoGPT for speech-to-text

A speech-to-text system you can train for $25.

Tiny Audio connects a frozen, pretrained speech encoder to a pretrained LLM with a small trainable projector. The model published from this repo gets 1.8% WER on LibriSpeech test-clean and 7.4% across 12 benchmarks (11,822 samples pooled) while training only ~80M parameters. The codebase is small enough to read in an afternoon, and you can run a training loop on your laptop in about five minutes.

No install: open the live demo, record yourself or upload a file, and get a transcript.

In Python:

pip install "transformers>=5.0" peft torch torchaudio librosa
python
from transformers import pipeline

pipe = pipeline(
    "automatic-speech-recognition", model="mazesmazes/tiny-audio", trust_remote_code=True
)
print(pipe("audio.wav")["text"])

The output is punctuated, capitalized, and has numbers formatted, with no post-processing step. The input can be a file path, a URL, or a 16 kHz numpy array. Weights are bf16, so you need roughly 6 GB of GPU or Apple Silicon memory.

pipe("audio.wav", return_timestamps=True)

pipe("meeting.wav", return_speakers=True, num_speakers=2)

Each speaker Nemotron-3-Diarization finds is transcribed separately, on a copy of the audio where everyone else is silenced, and each word belongs to the stream it came from. This is a zero-shot port of NeMo's masked_asr recipe; a word two streams both heard at once is kept once. Each speaker costs roughly their own talk time in ASR, and single-speaker audio is transcribed unmasked. Overlap is only partly handled: another person's speech inside a speaker's turn stays in that speaker's stream.

Speaker diarization needs transformers installed from main (pip install git+https://github.com/huggingface/transformers) until the next release. For token-by-token streaming output, see ASRModel.generate_streaming. The model card covers batching and GPU settings.

ta serve puts the model behind a batched HTTP server: requests arriving together share GPU batches, so throughput grows with load (about 460x real time at 128 concurrent requests on an RTX 4090). To run it on a RunPod GPU:

poetry run ta runpod up --serve                 # create an inference pod; prints <POD_ID>
poetry run ta runpod wait <POD_ID>              # prints <HOST> <PORT>
poetry run ta runpod deploy <HOST> <PORT>       # sync the project, install the fast kernels
TINY_AUDIO_API_KEY=my-secret poetry run ta runpod serve <HOST> <PORT> --no-attach

Without TINY_AUDIO_API_KEY the server is open to anyone who has the URL. ta serve also runs locally, on CUDA, Apple Silicon, or CPU.

Send the audio as the request body, with options in the query string:

curl -X POST "https://<POD_ID>-8000.proxy.runpod.net/?return_timestamps=true" \
  -H "Authorization: Bearer my-secret" \
  -H "Content-Type: application/octet-stream" \
  --data-binary @audio.wav
python
import httpx

response = httpx.post(
    "https://<POD_ID>-8000.proxy.runpod.net/",
    params={"return_speakers": "true", "num_speakers": "2"},
    content=open("meeting.wav", "rb").read(),
    headers={"Authorization": "Bearer my-secret"},
    timeout=600,
)
print(response.json()["text"])

• Options:return_timestamps ,return_speakers ,num_speakers andmax_speakers , as in the pipeline. The response is the same dict the pipeline returns.
• JSON body: to send JSON instead, use{"inputs": "<base64 audio>", "parameters": {...}} .
• Audio formats: anything FFmpeg can read.
• Errors:400 with{"error": ...} for bad audio or options, and401 for a wrong key.
• Other endpoints:GET /health andGET /stats (batch sizes and GPU time).

RunPod's HTTP proxy rejects request bodies over 500 MiB, and it drops any request that takes more than 100 seconds. For long recordings, send 16 kHz mono FLAC:

ffmpeg -i recording.wav -ac 1 -ar 16000 recording.flac

That's about 1 MB per minute of audio, and it costs nothing in accuracy, because the server converts everything to 16 kHz mono anyway. On an RTX 4090, 45 minutes of audio takes about 10 seconds, or 21 seconds with speaker labels. The demo Space calls the server this way.

Word error rate (%, lower is better) on 11,822 samples (up to 1,000 per dataset), measured with this repo's ta eval against the ta serve HTTP API on an RTX 4090:

† Held out: no data from this source was used in training.

You can check these numbers yourself and compare against commercial APIs on the same samples:

poetry run ta eval -m mazesmazes/tiny-audio -d loquacious -n 100
ASSEMBLYAI_API_KEY=... poetry run ta eval -m assemblyai -d loquacious -n 100
Audio (16 kHz) → speech encoder (frozen) → MLP projector (trained) → LLM decoder → Text

1. A pretrained speech encoder turns audio into a sequence of frame embeddings.
2. A small MLP projector stacks neighbouring frames and maps them into the LLM's embedding space. It is the only part trained from scratch.
3. The LLM reads those projected frames as if they were tokens and writes out the transcript.

Encoder, projector, and decoder are each swappable from config. Two recipes ship with the repo:

Start on your laptop for free, and rent a GPU only once you know the pipeline works.

git clone https://github.com/alexkroman/tiny-audio.git && cd tiny-audio
poetry install

poetry run python scripts/train.py +experiments=mps_smoke

poetry run ta runpod plan -e stage_1

poetry run python scripts/train.py +experiments=stage_1

Every setting is a Hydra override, for example model.projector_hidden_dim=2048 or training.use_lora=true. When you're happy with a model, ta push publishes it to the Hugging Face Hub and ta deploy puts a demo like the one above on a Space.

poetry run ta runpod up -e stage_1              # create a pod with enough GPU for the config
poetry run ta runpod wait <POD_ID>              # prints <HOST> <PORT>
poetry run ta runpod deploy <HOST> <PORT>       # sync the project and install dependencies
HF_TOKEN=hf_... poetry run ta runpod train <HOST> <PORT> -e stage_1
poetry run ta runpod attach <HOST> <PORT>       # watch the run in tmux

The free 3.5-hour course walks you through the full loop: how the encoder, projector, and decoder fit together (with real tensor shapes), training a model, evaluating it against commercial APIs, and publishing it with a live demo. You need Python, the command line, and git. The course trains the smaller stage_1 recipe, not the published model, so your WER will be higher than the table above.

Want to try a new projector architecture, add a dataset, or change the codebase? See CONTRIBUTING.md for the CLI reference, config layout, and quality gates.

• Granite Speech andGLM-ASR for audio encoding
• Qwen3.5 andQwen3 for language modeling
• LoquaciousSet for the default evaluation set

MIT