Greek Speech Synthesis Model

CSM 1B Adapted for Greek Language
HuggingFace Model HuggingFace Demo Arxiv Paper
Moira.AI
Abstract. This report documents the process of adapting the CSM 1B conversational speech model for the Greek language. The primary motivation was to address the scarcity of high-quality, open- source Text-to-Speech (TTS) models for Greek. By leveraging Low-Rank Adaptation (LoRA), a parameter-efficient finetuning technique, we successfully retrained the model on a curated dataset of high-quality Greek speech. The resulting model generates highly intelligible and natural-sounding Greek speech with clear articulation. This work not only provides a valuable open-source TTS tool for the Greek-speaking community but also validates LoRA as a resource- efficient and effective method for adapting large speech models to less-resourced languages.

Audio Samples

Greek Speech Generation

Sample audio generations demonstrating the natural and intelligible Greek speech produced by our adapted CSM 1B model.

Greek Text English Translation Audio Sample
Αγόρασα μήλα, πορτοκάλια, και μπανάνες, αλλά ξέχασα το γάλα. I bought apples, oranges, and bananas, but I forgot the milk.
Άσπρη πέτρα ξέξασπρη κι απ' τον ήλιο ξεξασπρότερη. White stone very white and from the sun even whiter.
Έκανα post στο Instagram και μετά πήγα για shopping στο mall. I posted on Instagram and then went shopping at the mall.
Η εκστρατεία προσέλκυσε χιλιάδες εθελοντές. The campaign attracted thousands of volunteers.
Ο παπάς ο παχύς, έφαγε παχιά φακή. Γιατί παπά παχύ, έφαγες παχιά φακή; The fat priest ate thick lentils. Why fat priest, did you eat thick lentils?

Voice Quality Demonstrations

Examples highlighting the clarity, naturalness, and articulation quality of the generated Greek speech across different sentence types and complexities.

Text Type Greek Example Audio Sample
Simple Statement Ο νόμος είναι σαφής.
Complex Sentence Ο λογαριασμός της Δ.Ε.Η. πρέπει να πληρωθεί, π.χ. μέσω τραπέζης.
Question Μια πάπια, μα ποια πάπια; Μια πάπια με παπιά.
Exclamation Ο βροντόσαυρος έψαχνε τροφή στην αυλή.

Long Form Generation

Extended text generation demonstrating the model's ability to maintain consistent voice quality and natural prosody over longer passages.

Greek Text English Translation Audio Sample
Κάθε πρωί, η ρουτίνα μου είναι απλή. Ξυπνάω συνήθως γύρω στις επτά, χωρίς ξυπνητήρι. Το πρώτο πράγμα που κάνω είναι να πιω ένα μεγάλο ποτήρι νερό. Μετά, ετοιμάζω έναν δυνατό καφέ, συνήθως έναν εσπρέσο, και κάθομαι για λίγα λεπτά στο μπαλκόνι. Μου αρέσει να κοιτάζω την πόλη καθώς ξυπνάει. Αφού πιω τον καφέ μου, κάνω ένα γρήγορο ντους και ετοιμάζομαι για τη δουλειά. Είναι ένας ήρεμος τρόπος για να ξεκινήσω τη μέρα μου. Every morning, my routine is simple. I usually wake up around seven, without an alarm clock. The first thing I do is drink a large glass of water. Then, I prepare a strong coffee, usually an espresso, and sit for a few minutes on the balcony. I like to watch the city as it wakes up. After drinking my coffee, I take a quick shower and get ready for work. It's a peaceful way to start my day.