🎼 Text2midi-InferAlign

Improving Symbolic Music Generation with Inference-Time Alignment

---

Text2midi-InferAlign is an inference-time technique that enhances symbolic music generation by improving alignment between generated compositions and textual prompts. It is designed to extend autoregressive models—like Text2Midi—without requiring any additional training or fine-tuning.

Our method introduces two lightweight but effective alignment-based objectives into the generation process:

• 🎵 Text-Audio Consistency: Encourages the temporal structure of the music to reflect the rhythm and pacing implied by the input caption.
• 🎵 Harmonic Consistency: Penalizes musically inconsistent notes (e.g., out-of-key or dissonant phrases), promoting tonal coherence.

By incorporating these alignment signals into the decoding loop, Text2midi-InferAlign produces music that is not only more faithful to textual descriptions but also harmonically robust.

We evaluate our technique on Text2Midi, a state-of-the-art text-to-MIDI generation model, and report improvements in both objective metrics and human evaluations.

---

📦 Installation & Usage

This repository contains the implementation of the Inference-Time Alignment module. Follow the steps below to get started.

1. Clone the Repository

git clone https://github.com/AMAAI-Lab/t2m-inferalign.git
cd t2m-inferalign

2. Set Up the Environment

We recommend using Python 3.10 and conda for environment management.

conda create -n alignment python=3.10
conda activate alignment
pip install -r requirements.txt

Please export your API key.

export ANTHROPIC_API_KEY=<your key>

or you can set your key here.

3. Download Model Weights and Resources

• Download the pretrained Text2Midi model from HuggingFace:
  🔗 https://huggingface.co/amaai-lab/text2midi

• Also download the corresponding tokenizer and soundfonts:
  🔗 https://huggingface.co/amaai-lab/text2midi/tree/main/

You may choose to organize them like this:

t2m-inferalign/
├── checkpoints/
│   └── pytorch_model.bin
├── tokenizer/
│   └── vocab_remi.pkl
├── soundfonts/
│   └── soundfont.sf2

Please fix the soundfont path here or here.

4. Run Inference with Alignment

python progressive_explorer.py --caption "A gentle piano lullaby with soft melodies" --model_path checkpoints/pytorch_model.bin --tokenizer_path tokenizer/vocab_remi.pkl --output_path outputs/lullaby.mid

Optional arguments:

• --max_tokens: Max number of tokens in the generated sequence.
• --batch_size: Number of tokens to generate before checking rewards.
• --beams: Number of parallel sequences to generate.

---

📊 Experimental Results

✅ Objective Evaluation

We evaluate on the MidiCaps dataset using six standard metrics. Our approach outperforms the Text2Midi baseline in all key alignment and tonal consistency metrics.

Metric	Text2Midi	Text2midi-InferAlign
CR (Compression Ratio) ↑	2.16	2.31
CLAP (Text-Audio Consistency) ↑	0.17	0.22
TB (Tempo Bin %) ↑	29.73	35.41
TBT (Tempo Bin w/ Tolerance %) ↑	60.06	62.59
CK (Correct Key %) ↑	13.59	29.80
CKD (Correct Key w/ Duplicates %) ↑	16.66	32.54

All results are averaged over ~50% of the MidiCaps test set (7913 captions randomly sampled).

---

🎧 Subjective Evaluation

A user study was conducted with 24 participants, comparing outputs from Text2Midi and Text2midi-InferAlign. Participants rated musical quality and text-audio alignment.

Music Quality & Text-Audio Match

Evaluation Criteria	Text2Midi (%)	Text2midi-InferAlign (%)
Music Quality	31.25	68.75
Text-Audio Match	41.67	58.33

Caption Type Preference

Caption Type	Text2Midi (%)	Text2midi-InferAlign (%)
MidiCaps Caption	48.33	51.67
Free Text Caption	27.78	72.22

These results demonstrate that Text2midi-InferAlign significantly enhances both musical structure and semantic relevance, especially for free-form, open-ended prompts.

---

📌 Citation

If you find this work useful in your research, please cite:

@article{text2midi-inferalign,
  title={Text2midi-InferAlign: Improving Symbolic Music Generation with Inference-Time Alignment},
  author={Abhinaba Roy, Geeta Puri, Dorien Herremans},
  year={2025},
  journal={arXiv:2505.12669}
}

---

🔗 Resources

• 🎧 Examples
• 🎼 Text2Midi (Base Model)
• 🤗 Text2Midi on HuggingFace