Robo-Shaul project

The Robo-Shaul project was a competition aimed at developing the first open Hebrew text-to-speech engine that replicates the voice of Shaul Amsterdamski, journalist & co-host of the popular Israeli podcast “Hayot Kis”. The competition ran from February, 2023 to May, 2023, and was documented in three episodes, two describing the process leading up to the competition and a final episode with the results.

Paper

The dataset and benchmark TTS system were published at INTERSPEECH 2023 as SASPEECH: A Hebrew Single Speaker Dataset for Text To Speech and Voice Conversion

Dataset

Available for download at OpenSLR

Code

Code & model for the benchmark TTS system is available on GitHub

Podcast Episodes

(Hebrew): Episode 1, Episode 2, Episode 3

Results

Transcript Ground Truth Baseline 1st Place 2nd Place 3rd Place 4th Place
הי, אתם על חיות כיס, אני שאול אמסטרדמסקי. אתם האזנתם לחיות כיס, הפודקאסט הכלכלי של כאן. עורך חיות כיס הוא תומר מיכלזון. במערכת חיות כיס תמצאו גם את צליל אברהם,
אבל מה שעובר על סמוטריץ הוא לא הסיפור. הסיפור הוא משהו אחר לגמרי.
ובשביל להבין למה מחיר הדלק כל כך עלה, צריך לחזור שנתיים אחורנית.
כששאלתי אותה למה היא החליטה לפתוח עסק של לק ג'ל, התשובה שלה היתה מאוד ברורה,
תגידו, גנבו לכם פעם את האוטו ופשוט ידעתם שאין טעם להגיש תלונה במשטרה?

People

Organizers

Shaul Amsterdamski - journalist & co-host of the Israeli podcast “Hayot Kis”

Orian Sharoni - ML researcher and founder of UpᐧAI consultancy

Roee Shenberg - ML researcher and founder of UpᐧAI consultancy

Uri Elyabayev - Consultant and founder of MDLI community

1st Place

Solution link (GitHub)

Max Melichov

As a versatile, analytical, and resourceful data scientist professional, I bring a wealth of experience in machine learning, data pre-processing, and voice processing techniques. With a proven track record in driving anomaly detection, disease detection, and deepfake algorithm construction, I am driven by the pursuit of organizational growth and thrive on triggering positive change. I champion efficiency, precision, and innovation, constantly seeking new opportunities to push the boundaries of data science.

LinkedIn, Email, Medium

Tony Hasson

A software dev with the goal of achieving expertise in Python, Data (analyzing + management),Automating Workflow and Machine Learning.

Besides that, I’m a problem solver, a great teacher, “out of the box” thinker, a people-person, mostly self taught, philomath, good with numbers and an outer-space geek.

LinkedIn

2nd Place

Solution link (GitHub)

Maxim Lvov

A researcher and algorithms developer in fields of AI, speech recognition and NLP.

LinkedIn

3rd Place

Solution link (GitHub)

Gabriel Khvalevsky

I am an independent and fast learner with a “can do” approach

You don’t require the perfect approval to commence your journey. I didn’t originate from the realm of artificial intelligence, nor did I undergo structured programming training. Nevertheless, I attempted, and I made multiple attempts using diverse approaches. Ultimately, I triumphed, and I take great pride in my accomplishments. In conclusion, I summarize my experience in the finest contest with this phrase: “Success greets those who persistently knock on the door.”

LinkedIn, Email

4th Place

Solution link (Google Colab)

Dr. Eyal Gruss

Code/media/text artist, algorithms researcher, teaches computational creativity at the Holon Institute of Technology.

Contact

For inquiries email us at roboshaul@upai.dev