Sayvors Models

Our Models for speech

Arabic-first models for text-to-speech and speech-to-text — built and fine-tuned on Gulf Arabic, each with its own codename and benchmark.

MModelSayvors · v2.1
MModelSayvors · v2
TTurtleSayvors Voice · v1
MModelSayvors · v2
MModelSayvors · v1
Text-to-Speech

TTS models

Voice generation models that turn text into natural Gulf-Arabic speech. Each has its own codename and benchmark.

M
ModelSayvors · v2.1
Current

Our fastest, most natural Gulf-Arabic voice — tuned for live phone conversations.

MOS (naturalness)4.8
Real-time factor (RTF)0.18
Arabic EPW — Gulf< 5%
Arabic EPC< 4.5%
Gulf dialect coverageHigh
M
ModelSayvors · v2
Stable

Balanced quality and speed for high-volume outbound calling.

MOS (naturalness)4.6
Real-time factor (RTF)0.21
Arabic EPW — Gulf< 6%
Arabic EPC< 5.5%
Gulf dialect coverageHigh
T
TurtleSayvors Voice · v1
Legacy

The original voice model — slower, with simpler prosody.

MOS (naturalness)4.3
Real-time factor (RTF)0.27
Arabic EPW — Gulf< 8%
Arabic EPC< 7%
Gulf dialect coverageHigh
Speech-to-Text

STT models

Transcription models that turn spoken Arabic into text in real time. Each has its own codename and benchmark.

M
ModelSayvors · v2
Current

Latest transcription model with strong Gulf-dialect accuracy.

WER — English2.4%
WER — Arabic Gulf4.8%
EPC — Arabic3.9%
Inference latency180 ms
Gulf dialectHigh
M
ModelSayvors · v1
Legacy

First-generation transcription baseline.

WER — English3.6%
WER — Arabic Gulf8.2%
EPC — Arabic6.7%
Inference latency340 ms
Gulf dialectHigh