METHOD, DEVICE AND MEDIA FOR ENCODING AND DECODING BIT STREAMS OF AUDIO DATA AND ASSOCIATED INFORMATION OF THE RETURN CHANNEL

Claims

1. A method for generating an encoded bit stream, carried out by a device with one or more microphones, wherein the method comprises the steps of: capture one or more audio signals using one or more microphones; analyze captured audio signals to determine the presence of the activation word; when the presence of an activation word is detected: set a flag to indicate that a speech recognition task should be performed on the captured audio signals: encode captured audio signals; collect the encoded audio signals and flag into an encoded bitstream.

2. The method according to claim 1, wherein one or more microphones are configured to capture a mono- or spatial sound field.

3. The method according to claim 2, wherein the spatial sound field has an A-format or a B-format.

4. The method according to any one of claims 1-3, wherein the captured audio signals are intended for use only in performing a speech recognition task.

5. The method of claim 4, wherein the captured audio signals are encoded such that when the captured audio signals are decoded, the quality of the decoded audio signals is sufficient to perform the speech recognition task, but is not sufficient for human listening.

6. The method according to claim 4 or 5, wherein the captured audio signals are transformed into a representation comprising one or more of frequency band energies, mel-frequency cepstral coefficients, or spectral coefficients of a modified discrete cosine transform (MDCT) prior to encoding the captured audio signals.

7. The method according to any one of paragraphs 1-3, in which the captured audio signals are intended both for human listening and for use in performing a speech recognition task.

8. The method of claim 7, wherein the captured audio signals are encoded such that when the captured audio signals are decoded, the quality of the decoded audio signals is sufficient for human listening.

9. The method according to claim 7, wherein encoding the captured audio signals comprises a step of generating a first encoded representation of the captured audio signals and a second encoded representation of the captured audio signals, wherein the first encoded representation is generated in such a way that when the captured audio signals are decoded from the first encoded representation, the quality of the decoded audio signals is sufficient for human listening, and wherein the second encoded representation is generated in such a way that when the captured audio signal signals are decoded from the second encoded representation, the quality of the decoded audio signals is sufficient for performing the speech recognition task, but is not sufficient for human listening.

10. The method of claim 9, wherein generating the second encoded representation of the captured audio signals comprises the step of converting the captured audio signals into one or more of a parametric representation, an approximate representation of the waveform, or a representation comprising one or more of frequency band energies, mel-frequency cepstral coefficients, or spectral coefficients of a modified discrete cosine transform (MDCT), prior to encoding the captured audio signals.

11. The method according to claim 9 or 10, wherein assembling the encoded audio signals into a bitstream comprises a step of inserting a first encoded representation into a first independent block of the encoded bitstream and inserting a second encoded representation into a second independent block of the encoded bitstream.

12. The method according to claim 9 or 10, wherein the first encoded representation is included in a first layer of the encoded bitstream, the second encoded representation is included in a second layer of the encoded bitstream, and the first and second layers are included in one block of the encoded bitstream.

13. The method according to any one of paragraphs 1-12, further comprising, when the presence of the wake word is not detected, the steps of: set a flag to indicate that the captured audio signals should not be subject to speech recognition; encode captured audio signals; collect the encoded audio signals and flag into an encoded bitstream.

14. A method for decoding an audio signal, comprising the steps of: receive an encoded bit stream containing encoded audio signals and a flag indicating whether a speech recognition task should be performed; decoding the encoded audio signals to obtain decoded audio signals; and when the flag indicates that the speech recognition task should be performed, perform the speech recognition task on the decoded audio signals.

15. The method of claim 14, wherein the decoded audio signals are intended to be used only in performing a speech recognition task.

16. The method according to claim 15, wherein the quality of the decoded audio signals is sufficient to perform the speech recognition task, but is not sufficient for human listening.

17. The method according to claim 15 or 16, wherein the decoded audio signals are in a representation comprising one or more of frequency band energies, mel-frequency cepstral coefficients, or spectral coefficients of a modified discrete cosine transform (MDCT) prior to encoding the captured audio signals.

18. The method of claim 14, wherein the captured audio signals are encoded such that when the captured audio signals are decoded, the quality of the decoded audio signals is sufficient for human listening.

19. The method according to claim 18, wherein the encoded audio signals comprise a first encoded representation of one or more audio signals and a second encoded representation of one or more audio signals.

20. The method according to claim 18, wherein the quality of the audio signals decoded from the first representation is sufficient for human listening, and wherein the quality of the audio signals decoded from the second representation is sufficient for performing the speech recognition task, but is not sufficient for human listening.

21. The method according to claim 19 or 20, wherein the first representation is in a first independent block of the coded bitstream, and the second representation is in a second independent block of the coded bitstream.

22. The method according to claim 19 or 20, wherein the first representation is in a first layer of the coded bitstream, the second coded representation is included in a second layer of the coded bitstream, and the first and second layers are included in one block of the coded bitstream.

23. The method according to any one of claims 18-22, wherein decoding the encoded audio signals comprises a step in which only the second representation is decoded and the first representation is ignored.

24. The method according to any one of claims 18-23, wherein the audio signals decoded from the second encoded representation are in a parametric representation, a waveform representation, or a representation comprising one or more of frequency band energies, mel-frequency cepstral coefficients, or spectral coefficients of a modified discrete cosine transform (MDCT).

25. A device capable of implementing the method according to any of paragraphs 1-24.

26. A non-transitory machine-readable storage medium containing a sequence of instructions that, when executed, cause one or more devices to perform the method according to any one of paragraphs 1-24.