sample-speech-short.wav (00:04 approx):
  "This sample voice recording is used for testing speech recognition systems."

sample-speech-long.wav / sample-speech-noisy.wav (same script, approx 00:35):
  "This long form speech sample was generated for testing transcription pipelines
   end to end. It contains several sentences with varied vocabulary, numbers like
   forty two and nineteen eighty four, and common technical words such as audio,
   sample rate, waveform, and transcript. A good speech to text test file should
   exercise punctuation, pauses, and longer utterances, because real world
   recordings are rarely short or clean. If your model can transcribe this
   paragraph accurately, your pipeline handles sustained input, not just single
   words."

License: CC0 1.0 Universal. Voice synthesized locally with eSpeak NG 1.51
(formant synthesis). Text and audio generated by SampleFiles — no third-party
recordings, no human voice, no attribution required.