Echo Clones 4M En: 4 Million Synthetic English Speech Utterances

SynData-2's echo-clones-4m-en dataset contains approximately 4 million English speech utterances generated for text-to-speech cloning. The data was synthesized using the EchoTTS model (jordand/echo-tts-base) and is based on 4,000 reference speakers. The dataset was last updated on HuggingFace on 2026-05-31.

Use Cases

• Training TTS models based on the 4 million synthetic utterances.
• Benchmarking voice cloning systems based on the 4,000 reference speakers.
• Analyzing synthetic speech quality based on the described text length categories (quip, mid, ramble).

Strengths

• Contains approximately 4 million audio samples.
• Uses a structured set of 4,000 reference speakers.
• Audio files have a consistent 44,100 Hz sample rate and 16-bit PCM format.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific TTS model and source text used for synthesis.

Provenance

Source

SynData-2 on HuggingFace.

Collection Method

Synthesized using the EchoTTS model (jordand/echo-tts-base).

Freshness

Last updated 2026-05-31 12:42:44; freshness should be verified.