Sphinx-4 1.0 Beta 6

Sphinx-4 1.0 Beta 6 appears to be a simple and efficient state-of-the-art speech recognition system that has been written entirely in the Java programming language. Sphinx-4 was created via a joint collaboration between the Sphinx group at Carnegie Mellon University, Mitsubishi Electric Research Labs (MERL), Sun Microsystems Laboratories, and Hewlett Packard (HP), with contributions from the University of California.

Sphinx-4 started out as a port of Sphinx-3 to the Java programming language, but evolved into a recognizer designed to be much more flexible than Sphinx-3, thus becoming an excellent platform for speech research. Sphinx-4 is a very flexible system capable of performing many different types of recognition tasks. As such, it is difficult to characterize the performance and accuracy of Sphinx-4 with just a few simple numbers such as speed and accuracy. 

Instead, they regularly run regression tests on Sphinx-4 to determine how it performs under a variety of tasks. These tasks and their latest results are as follows (each task is progressively more difficult than the previous task): 

Isolated Digits (TI46): Runs Sphinx-4 with pre-recorded test data to gather performance metrics for recognizing just one word at a time. The vocabulary is merely the spoken digits from 0 through 9, with a single utterance containing just one digit. (TI46 refers to the "NIST CD-ROM Version of the Texas Instruments-developed 46-Word Speaker-Dependent Isolated Word Speech Database".) 

Connected Digits (TIDIGITS): Extends the Isolated Digits test to recognize more than one word at a time (i.e., continuous speech). The vocabulary is merely the spoken digits from 0 through 9, with a single utterance containing a sequence of digits. (TIDIGITS refers to the "NIST CD-ROM Version of the Texas Instruments-developed Studio Quality Speaker-Independent Connected-Digit Corpus".) 

• Small Vocabulary (AN4): Extends the vocabulary to approximately 100 words, with input data ranging from speaking words as well as spelling words out letter by letter. 
• Medium Vocabulary (RM1): Extends the vocabulary to approximately 1,000 words. 
• Medium Vocabulary (WSJ5K): Extends the vocabulary to approximately 5,000 words. 
• Medium Vocabulary (WSJ20K): Extends the vocabulary to approximately 20,000 words. 
• Large Vocabulary (HUB4): Extends the vocabulary to approximately 64,000 words. 

Major Features:

1. Live mode and batch mode speech recognizers, capable of recognizing discrete and continuous speech.
2. Generalized pluggable front end architecture. Includes pluggable implementations of preemphasis, Hamming window, FFT, Mel frequency filter bank, discrete cosine transform, cepstral mean normalization, and feature extraction of cepstra, delta cepstra, double delta cepstra features.
3. Generalized pluggable language model architecture. Includes pluggable language model support for ASCII and binary versions of unigram, bigram, trigram, Java Speech API Grammar Format (JSGF), and ARPA-format FST grammars.
4. Generalized acoustic model architecture. Includes pluggable support for Sphinx-3 acoustic models.
5. Generalized search management. Includes pluggable support for breadth first and word pruning searches.
6. Utilities for post-processing recognition results, including obtaining confidence scores, generating lattices and embedding ECMAScript into JSGF tags.
7. Standalone tools. Includes tools for displaying waveforms and spectrograms and generating features from audio.