Dograh
Experiments  /  TTS verbalization benchmark  /  parler-tts
Model · EXP-001

parler-tts

Ranked 7 of 7 on the TTS verbalization benchmark. Strongest on abbreviation, weakest on version numbers.

At a glance
Said right3%share of all 1,080 sentences
Rank7 of 7among the seven models
Cost per clip3.06scomputing time per audio clip
Likely range2%–4%where the true score most likely sits
Strengths and weak spots

Where parler-tts does best and worst

Its five strongest and five weakest types of text, by the share of sentences it said right.

Best at
Abbreviation
0.57
Initialism or Acronym
0.15
Stock Ticker
0.03
Address
0.00
Biological Classification
0.00
Breaks on
Version Numbers
0.00
Vehicle or Product Code
0.00
Unit
0.00
URL or Email
0.00
Time
0.00
A real miss

Version Numbers: what parler-tts actually said

The text
This bug was fixed in v21.6.26 of the shared library.
Should be said as
This bug was fixed in version twenty one point six point twenty six of the shared library.
parler-tts said
"this bug was fixed in faplica busdray of the shared library"
All 27 types

Every type of text, best to worst

The share of sentences parler-tts said right on each of the 27 types. Each type has 40 test sentences.

Abbreviation
0.57
Initialism or Acronym
0.15
Stock Ticker
0.03
Address
0.00
Biological Classification
0.00
Cardinal
0.00
Chemical Formula
0.00
Currency
0.00
Date
0.00
Decimal
0.00
Fractions
0.00
Geographic Coordinates
0.00
Hashtag or Mention
0.00
ISBN
0.00
Legal Reference
0.00
License Plate or Serial Numbers
0.00
Mathematical Expression
0.00
Musical Notation
0.00
Ordinal
0.00
Phone Number
0.00
Roman Numeral
0.00
Sports score
0.00
Time
0.00
URL or Email
0.00
Unit
0.00
Vehicle or Product Code
0.00
Version Numbers
0.00