Stories
Slash Boxes
Comments

SoylentNews is people

posted by martyb on Monday March 01 2021, @03:15AM   Printer-friendly
from the do-you-hear-what-I-hear? dept.

Lyra: A New Very Low-Bitrate Codec for Speech Compression

Since the inception of Lyra, our mission has been to provide the best quality audio using a fraction of the bitrate data of alternatives. Currently, the royalty-free open-source codec Opus, is the most widely used codec for WebRTC-based VOIP applications and, with audio at 32kbps, typically obtains transparent speech quality, i.e., indistinguishable from the original. However, while Opus can be used in more bandwidth constrained environments down to 6kbps, it starts to demonstrate degraded audio quality. Other codecs are capable of operating at comparable bitrates to Lyra (Speex, MELP, AMR), but each suffer from increased artifacts and result in a robotic sounding voice.

Lyra is currently designed to operate at 3kbps and listening tests show that Lyra outperforms any other codec at that bitrate and is compared favorably to Opus at 8kbps, thus achieving more than a 60% reduction in bandwidth. Lyra can be used wherever the bandwidth conditions are insufficient for higher-bitrates and existing low-bitrate codecs do not provide adequate quality.

[...] The implications of technologies like Lyra are far reaching, both in the short and long term. With Lyra, billions of users in emerging markets can have access to an efficient low-bitrate codec that allows them to have higher quality audio than ever before. Additionally, Lyra can be used in cloud environments enabling users with various network and device capabilities to chat seamlessly with each other. Pairing Lyra with new video compression technologies, like AV1, will allow video chats to take place, even for users connecting to the internet via a 56kbps dial-in modem.

This should help make an 8 MiB copy of Shrek sound even better.

Also at CNX Software and Phoronix.


Original Submission

Related Stories

Google Posts First Beta Code for Lyra Speech Compression Codec 16 comments

Google Posts Initial Code For Lyra Speech Codec

Back in February we covered Google's work on the Lyra voice/audio codec designed for fitting with very low bit-rate audio for speech compression in use-cases like WebRTC and video chatting even on the most limited Internet connections. Thanks to leveraging machine learning, Lyra can function at just 3kbps. The code to Lyra is now public.

[...] The Lyra high-quality, low-bitrate speech codec is open-source with an initial v0.0.1 beta commit made today. Building Lyra requires the Bazel build system as well as a particular revision of LLVM/Clang for ABI compatibility.

GitHub (Apache-2.0 License).

Also at VentureBeat and CNX Software.

Previously:
Google Unveils Lyra Audio Codec with Better Speech Compression than Opus


Original Submission

Google Announces Lyra V2 Low Bit-Rate Speech Codec 12 comments

Google Announces Lyra V2 Low Bit-Rate Voice Codec

Lyra V2 is summed up by Google as being "a better, faster, and more versatile speech codec...a new architecture that enjoys a wider platform support, provides scalable bitrate capabilities, has better performance, and generates higher quality audio."

Lyra V2 makes use of the SoundStream end-to-end neural audio codec, continues showing much better performance than the Opus audio codec, improved audio quality, and more. The Lyra V2 open-source code is available today.

Lyra 1.2.0 on GitHub. New features:

  • Speed is significantly faster (~5x improvement seen on Android devices).
  • The SoundStream-based model produces significantly higher quality speech (when comparing 3kbps V1 to 3.2 kbps V2).
  • Selectable bitrate (3200, 6000, 9200 bits per second).
  • Codec latency reduced from 100 ms to 20 ms.
  • Mac and Windows support (in addition to continuing support for Linux and Android). Note: we have verified that these build, and run correctly, but have numerous compilation and linker warnings (Windows in particular due to MSVC/gcc mismatch). These issues and support for other platforms like iOS can be addressed by modifying the .bazelrc file. We welcome community contributions for this.
  • More portable code: The TensorFlow Lite model in the .tflite files can be used in other platforms. The TFLite runtime is optimized for individual platforms, replacing the need to write platform specific assembly.

Lyra (codec).

Previously: Google Unveils Lyra Audio Codec with Better Speech Compression than Opus
Google Posts First Beta Code for Lyra Speech Compression Codec


Original Submission

This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
(1)
  • (Score: 0) by Anonymous Coward on Monday March 01 2021, @04:08AM (2 children)

    by Anonymous Coward on Monday March 01 2021, @04:08AM (#1118377)

    The old telephone transmitted voice signal at analog 4khz band.

    Explain to me how it compares to this 3kps band.

    • (Score: 2, Informative) by pTamok on Monday March 01 2021, @09:20AM

      by pTamok (3042) on Monday March 01 2021, @09:20AM (#1118422)

      A standard uncompressed voice channel uses PCM (Pulse Code Modulation [wikipedia.org]) with 8-bit samples. The standard for analogue telephone bandwidth [wikipedia.org] is 4 kHz, which is sampled at 8,000 times per second (the Nyquist rate [wikipedia.org]), so a standard digitalised telephone voice channel, with no compression is 8,000 8-bit samples per second, which is 64,000 bits per second, or 64 kbit/s. In the USA, 1-bit was used for in-band signalling (robbed-bit signalling [wikipedia.org]), leading to standard digitalised voice channels being effectively 56 kbit/s for the audio, with a corresponding drop in quality.

    • (Score: 0) by Anonymous Coward on Monday March 01 2021, @03:07PM

      by Anonymous Coward on Monday March 01 2021, @03:07PM (#1118461)

      It's 1 better. It goes up to 4.

  • (Score: 0) by Anonymous Coward on Monday March 01 2021, @04:24AM

    by Anonymous Coward on Monday March 01 2021, @04:24AM (#1118380)

    Civilian consumer stuff, even in crummy countries, isn't so constrained.

    Police and fire departments might contemplate it, at least in calm conditions, for centralized encrypted communication. (but might as well use ordinary cell networks for calm conditions) In a disaster situation, unencrypted peer-to-peer AM is a far better bet for reliable communication.

    The military might like it. They often use HF links, trying to allow for a war in which all satellites are nuked. HF is low-bandwidth, and encryption would be highly desirable.

  • (Score: 0) by Anonymous Coward on Monday March 01 2021, @05:42AM

    by Anonymous Coward on Monday March 01 2021, @05:42AM (#1118388)

    s/t

  • (Score: 3, Insightful) by fakefuck39 on Monday March 01 2021, @06:00AM (5 children)

    by fakefuck39 (6620) on Monday March 01 2021, @06:00AM (#1118390)

    Umm, yeah, when you compare it to a general purpose audio compressor, of course - what an amazing breakthrough. But those aren't used to compress voice. Funny how google gives the Opus (used in the wrong mode) bitrate, then mentions other codecs but doesn't give their bitrate (which is comparable to google's).

    DoD's CELP, speex, etc, are all under 5kb/s, which is what this should be compared to. Heck, CDMA2000, the 3G technology America was using a long time ago when the rest of the world was stuck with 2G GSM, used VMR (designed by Nokia) - about 4-6kb/s. I'll just assume later tech is even better. Strange, I don't remember my phone sounding "robotic" unless I get real crappy reception.

    >Opus can be used in more bandwidth constrained environments down to 6kbps
    actually it's 5kb/s, because it has a narrowband mode, which is what should be used for voice, so not only is the comparison to Opus in the article a strawman, it even (purposely?) compares it to the higher bandwidth full band mode in Opus.

    Why? Oh, because this is a story published by google, to make their non-groundbreaking tech that is not any better than existing codecs, appear better than it is. Here's the real non-news: google products are all shine, and are buggy amateur crap you're beta testing.

    so, google. great work. you've given a 25% incremental improvement over voice codecs from a decade ago. good job, seriously. but nothing spectacular or ground breaking. just expected. just like google lying.

    • (Score: 3, Informative) by linuxrocks123 on Monday March 01 2021, @07:05AM (4 children)

      by linuxrocks123 (2557) on Monday March 01 2021, @07:05AM (#1118399) Journal

      They did compare it with SPEEX at 3kbps. SPEEX at 3kbps sounded very glitchy in their sample.

      Can you elaborate on what they did wrong with Opus? I used opusenc --bitrate 6 and got pretty much what they got.

      • (Score: 0) by Anonymous Coward on Monday March 01 2021, @02:13PM

        by Anonymous Coward on Monday March 01 2021, @02:13PM (#1118449)

        I can't answer the question, but I know that Opus is not a codec, it's a hybrid between the codecs SILK from Skype and CELT.

      • (Score: 2) by fakefuck39 on Tuesday March 02 2021, @01:45AM (2 children)

        by fakefuck39 (6620) on Tuesday March 02 2021, @01:45AM (#1118720)

        my point is they either compare full-band opus at 32 or speex at 3. speex at 5 would be the real comparison. but they don't do that test at all, because it doesn't fit the narrative. for opus narrowband, look ayt narrowband on the opus wiki.

        • (Score: 2) by linuxrocks123 on Tuesday March 02 2021, @03:51AM (1 child)

          by linuxrocks123 (2557) on Tuesday March 02 2021, @03:51AM (#1118751) Journal

          They did opus at 6, not opus at 32. After looking around, I don't think they did anything wrong with opus. Like I said, I did opusenc --bitrate 6 and got pretty much what they did. The encoder should be using the right algorithm automatically, I think.

          Why should they do speex at 5? If they're 3, the right comparison is anything at 3. opus only goes down to 6, though, at least opusenc only goes down to 6.

          • (Score: 2) by fakefuck39 on Tuesday March 02 2021, @05:36AM

            by fakefuck39 (6620) on Tuesday March 02 2021, @05:36AM (#1118781)

            so why didn't they use opys at 3 using your logic? oh, because they're claiming bitrate reduction for same quality. so that's why they'd use speex at 5. a reduction of 40%. and that's the right comparison. and they did opus at 32 and 8, it's right in the summary. you didn't read the opus wike, did you. it should have linked you to codec2. mostly because opus is a container, which uses several codecs. what it appears you did is use opus, the container, with a codec that's optimized for higher bitrate. both you and google tested transporting a piano in a sports car and brilliantly concluded your new truck works better.

  • (Score: 4, Insightful) by zeigerpuppy on Monday March 01 2021, @06:35AM

    by zeigerpuppy (1298) on Monday March 01 2021, @06:35AM (#1118394)

    did Google include the sentence about Opus being open-source so that people would think Lyra is too?
    The less we rely on google codecs, 'standards' and infrastructure, the better in my opinion.

  • (Score: 1) by pTamok on Monday March 01 2021, @08:11AM (2 children)

    by pTamok (3042) on Monday March 01 2021, @08:11AM (#1118414)

    The comments on the Phoronix article about Google's Lyra [phoronix.com] pointed me towards LPCNet:

    A Real-Time Wideband Neural Vocoder at 1.6 kb/s Using LPCNet [jmvalin.ca]

    The LPCNet source code is available under a BSD license.

  • (Score: 0) by Anonymous Coward on Monday March 01 2021, @07:23PM (2 children)

    by Anonymous Coward on Monday March 01 2021, @07:23PM (#1118570)

    [...] The implications of technologies like Lyra are far reaching, both in the short and long term. With Lyra, billions of users in emerging markets can the NSA will have access to an efficient low-bitrate codec that allows them to have store more, higher quality audio than ever before.

    • (Score: 0) by Anonymous Coward on Monday March 01 2021, @10:30PM (1 child)

      by Anonymous Coward on Monday March 01 2021, @10:30PM (#1118660)

      You're footing the bill. They will use high-bitrate FLAC.

(1)