How Computers Speak: Written Text, Spoken Words, and Speech Engines

Episodes

1. 1

   Written Text Becomes Speech

   You type a sentence and a voice reads it back. Discover that a speech system is a chain of stages, and this series starts at the front door: messy writing in, a sound file out.

2. 2

   From Symbols to Spoken Words

   A speech engine reads what you literally give it. Dollar signs, dates, and phone numbers have to become the words a person would actually say, in a careful order.

3. 3

   Three Engines, Old and New

   The same spoken-form sentences go to an offline engine and two neural engines. Listen for privacy, naturalness, and who is allowed to change pitch.

4. 4

   Storing Speech and Measuring Sound

   Every clip is stored the same way on purpose. Sample rate, bit depth, and an uncompressed wave are choices with reasons, and the waveform, frequency content, and loudness of each clip tell you what listening alone can miss.

5. 5

   What the Comparison Actually Showed

   Thirty spoken clips later, the ranking is not a surprise: the neural engines sound more natural. The surprise is why they speak more slowly, and why they still generate faster.