AI · Theory

Speech Recognition

Theory

169/215

Speech Recognition

Speech recognition: audio waveform → text. Steps: features (MFCCs or learned), acoustic model, language model (today often end-to-end nets). Noise, accents, and overlapping talk still hurt. Not the same as speaker identification (who spoke).

Output — idea / do / check for Speech Recognition. Fill those three in the viva.

Exam tip

ASR vs speaker ID.

Example

# Speech Recognition
print("idea :", "Speech Recognition")
print("do   : one tiny example on paper")
print("check: one failure case")

Output: idea / do / check for Speech Recognition. Fill those three in the viva.

Short notes

DefAudio → text.
Rule≠ speaker ID.

AI · Theory

Speech Recognition

← All stacks

Theory

169/215

Speech Recognition

Output — idea / do / check for Speech Recognition. Fill those three in the viva.

Exam tip

ASR vs speaker ID.

Example

# Speech Recognition
print("idea :", "Speech Recognition")
print("do   : one tiny example on paper")
print("check: one failure case")

Output: idea / do / check for Speech Recognition. Fill those three in the viva.

Short notes

DefAudio → text.
Rule≠ speaker ID.

Speech Recognition

Example

Short notes

Questions

Speech Recognition

Example

Short notes

Questions