Part-Time Audio AI Researcher — Student

About Besimple AI

Besimple AI is building data and evaluation infrastructure for the next generation of voice AI.

We work with AI labs and model companies to help them evaluate and improve speech and voice systems through high-quality datasets, rigorous benchmarks, and scalable evaluation workflows.

We recently released VoiceCodeBench, a benchmark that evaluates how accurately ASR systems recover structured information—such as email addresses, URLs, file paths, command-line arguments, phone numbers, dates, and measurements—from spoken workplace audio.

VoiceCodeBench:
https://huggingface.co/datasets/besimple-ai/voice-code-bench

We are looking for a part-time student researcher with hands-on research experience in audio, speech, or voice AI. Your primary focus will be evaluating state-of-the-art voice AI models, analyzing where they fail, and helping us develop new benchmarks and datasets.

What You’ll Do

1. Evaluate state-of-the-art ASR, TTS, speech-to-speech, and voice agent models.
2. Design experiments that compare models across accuracy, latency, robustness, accents, noise conditions, speaker variation, and other dimensions.
3. Perform detailed error analysis by reviewing audio, transcripts, model outputs, and evaluation metrics.
4. Help develop new audio benchmarks and datasets based on emerging model capabilities and customer needs.
5. Implement reproducible evaluation scripts and research workflows in Python.
6. Review recent audio AI research and identify promising models, datasets, metrics, and evaluation methods.
7. Summarize findings in clear research notes, benchmark reports, and technical documentation.
8. Work directly with the founders and technical team to turn research questions into concrete experiments.

What We’re Looking For

You are currently pursuing a bachelor’s, master’s, or PhD degree in computer science, machine learning, electrical engineering, computational linguistics, signal processing, or a related field.

You should have:

1. Research or project experience in audio AI, speech processing, ASR, TTS, voice agents, or a closely related area.
2. Experience working with speech models, audio datasets, or speech evaluation methods.
3. Strong Python skills and familiarity with common machine learning workflows.
4. Experience with at least some relevant tools, such as PyTorch, Hugging Face, Whisper, torchaudio, librosa, or ffmpeg.
5. The ability to independently design experiments, analyze results, and communicate findings.
6. Strong attention to detail, especially when reviewing audio samples and model errors.
7. Comfort working in a fast-moving startup environment with open-ended research questions.
8. Sufficient English proficiency to read research papers and write clear technical notes.
9. Prior full-time industry experience is not required. Strong coursework, academic research, lab experience, open-source contributions, or independent projects are all relevant.

Nice to Have

1. Experience evaluating ASR or TTS systems using metrics such as WER, CER, speaker similarity, latency, intelligibility, or task success rate.
2. Experience with multilingual speech, accents, noisy audio, long-form audio, or low-resource languages.
3. Experience publishing a paper, technical report, benchmark, dataset, or open-source project.
4. Familiarity with commercial voice AI APIs or open-source speech models.
5. Experience building evaluation pipelines or working with large audio datasets.
6. Participation in an academic speech, audio, NLP, or machine learning research lab.

Example Research Questions

Some of the questions you may work on include:

1. Which ASR models are best at recognizing structured information?
2. How does performance change across accents, speakers, microphones, and noise conditions?
3. Which evaluation metrics correlate with real-world voice agent performance?
4. Where do leading speech-to-speech systems fail during long or complex conversations?
5. How should we evaluate interruptions, turn-taking, latency, and task completion in voice agents?
6. What new datasets will be needed to evaluate the next generation of voice models?

Time Commitment

1. Part-time, approximately 10–20 hours per week
2. Flexible schedule around classes and research commitments
3. Initial commitment of 3–6 months, with the possibility of extension

Location: Remote

Compensation: $40 per hour

Why Join Besimple AI

1. Work on research problems directly connected to cutting-edge voice AI systems.
2. Build benchmarks and datasets that can influence how audio AI models are evaluated.
3. Gain hands-on experience across speech research, data, evaluation, and production workflows.
4. Work directly with the founders and take meaningful ownership of your projects.
5. Produce work that may lead to public benchmarks, datasets, technical reports, or open-source releases.
6. Join a small team where your research can quickly become part of a real product.

Interview Process

1. Introductory conversation
2. Discussion of a relevant research project, paper, or technical project you have worked on
3. Short practical exercise or research case involving audio AI evaluation
4. Final conversation with the founding team

How to Apply

Please include:

1. Your résumé or CV
2. A brief description of your experience in audio, speech, or voice AI
3. Links to relevant papers, GitHub repositories, datasets, demos, or course projects
4. Your expected weekly availability and preferred start date