Acoustic scoring unit implemented on a single FPGA or ASIC

• US 8,639,510 B1
• Filed: 12/22/2008
• Issued: 01/28/2014
• Est. Priority Date: 12/24/2007
• Status: Expired due to Fees

• Current Assignee: Carnegie Mellon University
• Original Assignee: Carnegie Mellon University
• Inventors: Rutenbar, Rob A., Yu, Kai
• Primary Examiner(s): PULLIAS, JESSE SCOTT
• Application Number:
  US12/341,160
• Time in Patent Office: 1,863 Days
• Field of Search: 704/231, 704/255, 704/256, 704/256.1, 704/256.4, 704/256.6, 704/256.7, 704/E15.05
• US Class Current: 704/256.6
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 2015/085 Methods for reducing search complexity, pruning

Abstract

A hardware acoustic scoring unit for a speech recognition system and a method of operation thereof are provided. Rather than scoring all senones in an acoustic model used for the speech recognition system, acoustic scoring logic first scores a set of ciphones based on acoustic features for one frame of sampled speech. The acoustic scoring logic then scores senones associated with the N highest scored ciphones. In one embodiment, the number (N) is three. While the acoustic scoring logic scores the senones associated with the N highest scored ciphones, high score ciphone identification logic operates in parallel with the acoustic scoring unit to identify one or more additional ciphones that have scores greater than a threshold. Once the acoustic scoring unit finishes scoring the senones for the N highest scored ciphones, the acoustic scoring unit then scores senones associated with the one or more additional ciphones.

Claims

1. A hardware implemented acoustic scoring unit for a speech recognition system, said acoustic scoring unit comprising a single application specific integrated circuit (ASIC) that includes:

control logic, implemented on the single ASIC, and configured to receive feature vectors indicative of acoustic input and pass said feature vectors to acoustic scoring logic, also implemented on the single ASIC;

senone selection logic, also implemented on the single ASIC and coupled to said control logic, said senone selection logic operating in parallel with said acoustic scoring logic to select senones for scoring by said acoustic scoring logic;

whereby the parallel operation of said senone selection logic and said acoustic scoring logic enables said acoustic scoring logic to substantially continuously perform acoustic scoring operations until its scoring for a frame in question is complete.

2. An acoustic scoring unit as defined in claim 1, wherein said senone selection logic includes:

a senone scoring queue, implemented as SRAM on the single ASIC, and coupled to the acoustic scoring logic.

3. An acoustic scoring unit as defined in claim 2, wherein said senone selection logic also includes:

queue insertion logic, implemented on the single ASIC, and coupled to the senone scoring queue.

4. An acoustic scoring unit as defined in claim 1, wherein the control logic is coupled to an external memory device, and wherein the single ASIC also includes:

an external memory address repository, implemented as SRAM on the single ASIC, and configured to store addresses used to address the external memory device, which contains constants used to score senones.

5. A hardware implemented acoustic scoring unit for a speech recognition system, said acoustic scoring unit comprising a single field programmable gate array (FPGA) that includes:

control logic, implemented on the single FPGA, and configured to receive feature vectors indicative of acoustic input;

acoustic scoring logic, implemented on the single FPGA, and coupled to the control logic and to an external memory device;

senone selection logic, also implemented on the single FPGA and coupled to said control logic, said senone selection logic operating in parallel with said acoustic scoring logic to select senones for scoring by said acoustic scoring logic;

whereby the parallel operation of said senone selection logic and said acoustic scoring logic enables said acoustic scoring logic to substantially continuously perform acoustic scoring operations until its scoring for a frame in question is complete.

6. An acoustic scoring unit as defined in claim 5, wherein the senone selection logic includes:

a senone scoring queue, implemented as SRAM on the single FPGA, and coupled to the acoustic scoring logic.

7. An acoustic scoring unit as defined in claim 6, wherein the senone selection logic also includes:

queue insertion logic, implemented on the single FPGA, and coupled to the senone scoring queue.

8. An acoustic scoring unit as defined in claim 5, wherein the control logic is coupled to an external memory device, and wherein the single FPGA also includes:

an external memory address repository, implemented as SRAM on the single FPGA, and configured to store addresses used to address the external memory device, which contains constants used to score senones.

9. A hardware-implemented acoustic scoring stage for a speech recognition system, said acoustic scoring stage comprising:

(1) an external memory device, which stores constants used to score senones; and

,

(2) a single application specific integrated circuit (ASIC) that includes at least;

(a) control logic, implemented on the single ASIC, and configured to receive feature vectors indicative of acoustic input;

(b) acoustic scoring logic, implemented on the single ASIC, and coupled to the control logic and to the external memory device;

(c) senone selection logic, also implemented on the single ASIC and coupled to said control logic, said senone selection logic operating in parallel with said acoustic scoring logic to select senones for scoring by said acoustic scoring logic;

whereby the parallel operation of said senone selection logic and said acoustic scoring logic enables said acoustic scoring logic to substantially continuously perform acoustic scoring operations until its scoring for a frame in question is complete.

10. A hardware-implemented acoustic scoring stage for a speech recognition system as defined in claim 9, wherein the senone selection logic includes:

a senone scoring queue, implemented as SRAM on the single ASIC, and coupled to the acoustic scoring logic.

11. A hardware-implemented acoustic scoring stage for a speech recognition system as defined in claim 10, wherein the senone selection logic also includes:

queue insertion logic, implemented on the single ASIC, and coupled to the senone scoring queue.

12. A hardware-implemented acoustic scoring stage for a speech recognition system as defined in claim 9, wherein the single ASIC further includes:

an external memory address repository, implemented as SRAM on the single ASIC, and configured to store addresses used to address the external memory device, which contains constants used to score senones.