In an October 2026 blind listening test, 160 kbps was the lowest Opus bitrate that reached strict perceptual transparency for music, meaning even the most critical listeners in the test could not reliably tell it from the uncompressed original (p = 0.024 for equivalence) (Deezer Research, Martin and Hennequin, arXiv 2026). That happens to be the exact audio bitrate OBS Studio sets by default on all six of its audio tracks (OBS Project, OBS Studio source code 2026), yet the live platforms ask for less: YouTube Live recommends 128 kbps for stereo (Google, YouTube Help 2026) and Meta prefers 128 kbps for Facebook Live (Meta for Developers, Live Video API reference 2026). At a 96 kbps target, Opus already scored 4.65 out of 5 in a public multiformat blind test (Hydrogenaudio, public listening test 2014). Unlike on-demand files, covered in our podcast hosting bandwidth and storage statistics, a live encoder gets one pass and no second chance. We aggregated data from Deezer Research, YouTube Help, Meta for Developers, the Kick Help Center, the OBS Project, the IETF, Apple, the EBU, ITU-R, Spotify, and the Hydrogenaudio listening-test community.
TL;DR
- Opus at 160 kbps was statistically equivalent to uncompressed audio (p = 0.024), but the 10 most critical listeners caught 128 kbps (p = 0.030) (Deezer Research, arXiv 2026).
- YouTube Live asks for 128 kbps stereo at 44.1 kHz, a third of the 384 kbps at 48 kHz it recommends for uploaded videos (YouTube Help, live and upload settings 2026).
- Facebook Live prefers 128 kbps AAC-LC stereo audio, with 256 kbps as the maximum (Meta for Developers, Live Video API reference 2026).
- OBS Studio’s built-in presets cap audio at 320 kbps for Twitch, 160 kbps for YouTube RTMPS and 128 kbps for Facebook Live (OBS Project, services.json 2026).
- OBS Studio defaults to 160 kbps audio, 48,000 Hz and stereo (OBS Project, OBS Studio source code 2026).
- Apple’s HLS spec caps stereo AAC at 160 kbit/s and says HE-AAC should not be used above 64 kbit/s (Apple, HLS Authoring Specification 2026).
- At a 96 kbps target, Opus scored 4.65 vs 4.40 for Apple AAC on a 1 to 5 scale (Hydrogenaudio, public listening test 2014).
- The IETF Opus standard recommends 28 to 40 kbit/s for fullband speech and 64 to 128 kbit/s for stereo music (IETF, RFC 6716 2012).
- Spotify streams podcasts at about 96 kbit/s in its apps (Spotify Support, Audio quality 2026).
- EBU guidance puts the last limiter before a lossy encoder at -2 dBTP (EBU, Tech 3344 v2.1 2016).
1. Platform Ingest Specs: What Each Service Asks For
“Ingest” means the upload leg: the stream your encoder sends to the platform before the platform re-encodes it for viewers. On that leg, the published audio numbers are strikingly modest. YouTube asks for 128 kbps stereo audio on a live stream but 384 kbps stereo on an uploaded video, three times as much (Google, YouTube Help 2026). Live audio is treated as a contribution feed that must survive a congested home uplink, not as a master.
The sample rate guidance does not line up either. YouTube Live lists 44.1 kHz for stereo, while its own upload page and the OBS default use 48 kHz, and Kick allows up to 48 kHz. That mismatch costs nothing audible on its own, but every resample between a microphone, a mixer and an encoder is a step where a wrong setting can creep in. Audio settings are also not the only audio rule a channel hits: rights management is the other, as our live stream copyright mute statistics show.
| Metric | Value | Source |
|---|---|---|
| YouTube Live audio: stereo vs 5.1 surround | 128 kbps at 44.1 kHz vs 384 kbps at 48 kHz (5.1 over RTMP or RTMPS only in AAC) | Google, YouTube Help: Choose live encoder settings (2026) |
| YouTube recommended audio for uploaded (not live) videos: mono / stereo / 5.1 | 128 / 384 / 512 kbps at 48 kHz | Google, YouTube Help: Recommended upload encoding settings (2026) |
| Facebook Live audio bitrate: preferred vs maximum | 128 kbps vs 256 kbps (AAC low complexity, stereo) | Meta for Developers, Live Video API reference (2026) |
| Facebook Live audio sample rates accepted | 44.1 kHz or 48 kHz | Meta for Developers, Live Video API reference (2026) |
| Kick audio limit | Stereo, maximum 48 kHz; no audio bitrate published | Kick Help Center, How to stream on Kick (2026) |
| Maximum audio bitrate in OBS Studio’s built-in service presets: Twitch / YouTube RTMPS / Trovo / Facebook Live | 320 / 160 / 160 / 128 kbps | OBS Project, services.json preset file (2026) |
Context note: Twitch’s own Broadcasting Guidelines page loads through a script that could not be read for this roundup, so the Twitch figure above is a preset ceiling from broadcast software, not a quote from Twitch. Third-party guides cite a lower Twitch ceiling (see the Data watch note in Methodology).
2. Audio’s Share of the Bitrate Budget
Audio is a rounding error on a 1080p stream and a major line item on a mobile one. At YouTube’s 0.4 Mbps minimum for 360p H.264, a 128 kbps stereo track equals 32% of the video bitrate (128 / 400), while at the 17 Mbps recommendation for 1080p60 it is about 0.75% (128 / 17,000) (Google, YouTube Help 2026). The math below is ours, from the published figures, and compares audio with the video bitrate alone.
The practical reading: on a high-resolution stream, cutting audio from 160 to 128 kbps saves almost nothing, so audio is the last place to economize. On the low rungs of an adaptive ladder, the ones phone viewers on weak networks actually receive, audio bits compete directly with picture, which is why adaptive streaming specs allow stereo AAC as low as 32 kbit/s and steer low rates toward HE-AAC.
| Metric | Value | Source |
|---|---|---|
| YouTube Live 1080p60 H.264 recommended video bitrate, and 128 kbps audio as a share of it | 17 Mbps; 128 / 17,000 = about 0.75% | Google, YouTube Help: Choose live encoder settings (2026) |
| YouTube Live 360p30 H.264 minimum video bitrate, and 128 kbps audio as a share of it | 0.4 Mbps; 128 / 400 = 32% | Google, YouTube Help: Choose live encoder settings (2026) |
| Facebook Live 1080p60 video bitrate range, and 128 kbps preferred audio as a share of it | 4,500 to 9,000 kbps; 128 / 9,000 = about 1.4%, 128 / 4,500 = about 2.8% | Meta for Developers, Live Video API reference (2026) |
| Kick video bitrate range (H.264 only, constant bitrate only) | 1,000 to 8,000 kbps | Kick Help Center, How to stream on Kick (2026) |
| OBS Studio Twitch preset: maximum video vs maximum audio bitrate | 6,000 vs 320 kbps; 320 / 6,000 = about 5.3% | OBS Project, services.json (2026) |
| Apple HLS stereo AAC bitrate range | 32 to 160 kbit/s | Apple, HLS Authoring Specification for Apple Devices (2026) |
| Apple HLS: bitrate above which HE-AAC should not be used | 64 kbit/s | Apple, HLS Authoring Specification (2026) |
| Apple HLS live streams: allowed one-hour average and peak vs declared bandwidth | under 110% of AVERAGE-BANDWIDTH; under 125% of BANDWIDTH | Apple, HLS Authoring Specification (2026) |
Context note: Kick requires constant bitrate (CBR), meaning the encoder sends a steady rate rather than spending more bits on hard scenes. CBR makes the audio share above fixed, so a stereo track at 160 kbps on an 8,000 kbps Kick stream is 2% of the video budget (160 / 8,000).
3. Defaults and Codec Plumbing: AAC In, Opus Arriving
Most streamers never touch the audio panel, so defaults are the real industry standard. OBS Studio ships 160 kbps on every audio track, at 48,000 Hz and in stereo (OBS Project, OBS Studio source code 2026). That single default sits above what YouTube and Facebook recommend and below the OBS ceiling for Twitch, which is a sensible middle when one encoder feeds several platforms. If your processed voice enters OBS from another app, as with a voice changer for OBS, matching that 48 kHz rate end to end avoids one extra resample.
The codec story is shifting underneath. Every RTMP ingest spec in section 1 names AAC, but the WebRTC world mandates Opus, and WHIP (the WebRTC-HTTP Ingestion Protocol, an HTTP-based way to push a WebRTC stream into a server) brings that world to broadcast software. OBS added WHIP output in version 30.0, and the IETF published WHIP as RFC 9725 in 2025, with an example offer that negotiates Opus at 48 kHz in stereo (“opus/48000/2”).
| Metric | Value | Source |
|---|---|---|
| OBS Studio default audio bitrate, Simple output and all six Advanced output tracks | 160 kbps | OBS Project, OBS Studio source code, InitBasicConfigDefaults (2026) |
| OBS Studio default sample rate and channel layout | 48,000 Hz, stereo | OBS Project, OBS Studio source code (2026) |
| Opus recommended bitrates with 20 ms frames: fullband speech vs fullband stereo music | 28 to 40 kbit/s vs 64 to 128 kbit/s | IETF, RFC 6716 (September 2012) |
| Opus fullband mode: effective sample rate and highest coded frequency | 48 kHz; audio above 20 kHz is never coded | IETF, RFC 6716 (September 2012) |
| Audio codecs every WebRTC endpoint must implement | Opus plus G.711 (PCMA and PCMU) | IETF, RFC 7874 (May 2016) |
| OBS Studio release that added WHIP/WebRTC output | 30.0, November 2023 | OBS Project, OBS Studio 30.0 release notes (2023) |
Context note: the WHIP standard itself names no mandatory codec; Opus appears only in its example exchange. The Opus requirement comes from the separate WebRTC codec rules in the table, which govern WebRTC endpoints generally.
4. AAC vs Opus in Blind Listening Tests
Blind tests are the only fair referee between codecs, and the public record is older than the streaming stack it judges. In the July 2014 Hydrogenaudio public test at a 96 kbps target, Opus 1.1 scored 4.65 and Apple AAC 4.40 on a 1 to 5 impairment scale, where 5 means “imperceptible” (Hydrogenaudio, public multiformat listening test 2014). The gap is small in absolute terms but statistically solid (p = 0.000), and it came from Opus spending only 3 kbps more on average (106.7 vs 103.7 kbps).
Most recent available data: 2010 to 2014 test results (sources in the table). No comparable public multiformat test of current AAC and Opus encoders at streaming bitrates was found for 2025 or 2026, which is itself a finding: the platforms’ AAC defaults rest on compatibility, not on fresh listening data. For the very low bitrates where neural codecs now compete with Opus, see our neural audio codec bitrate and quality statistics.
| Metric | Value | Source |
|---|---|---|
| Mean score at a 96 kbps target, 1 to 5 scale (5 = imperceptible): Opus 1.1 vs Apple AAC vs LAME MP3 vs Vorbis | 4.65 vs 4.40 vs 4.24 vs 4.24 | Hydrogenaudio, public multiformat listening test results (July 2014) |
| Actual average bitrate delivered at that target: Opus / Apple AAC / Vorbis / MP3 | 106.7 / 103.7 / 106.1 / 135.5 kbps | Hydrogenaudio, public multiformat listening test (2014) |
| Low-quality anchor (an older AAC encoder, FAAC, at 96 kbps) | 2.65 | Hydrogenaudio, public multiformat listening test (2014) |
| Significance: Opus vs Apple AAC; Apple AAC vs MP3 | Opus better, p = 0.000; AAC better, adjusted p = 0.043 (bootstrap) | Hydrogenaudio, public multiformat listening test (2014) |
| Panel and material | 38 participants, 339 valid results of 363 submitted; 40 samples (35 music, 5 speech) | Hydrogenaudio, public multiformat listening test (2014) |
| Google stereo music test (9 listeners): Opus at 128 kbps vs AAC-LC at 64 kbps, MP3 at 96 kbps, Opus at 64 kbps and G.719 at 64 kbps | Opus at 128 kbps scored significantly higher than the other four | IETF, draft-ietf-codec-results-03 (May 2013) |
| Broadcom stereo music test (17 listeners, December 2010): Opus at 96 kbps vs AAC-LC and MP3 | Slightly worse than AAC-LC, significantly better than MP3 | IETF, draft-ietf-codec-results-03 (May 2013) |
Outlier note: MP3 used 28.8 kbps more than Opus (135.5 minus 106.7 kbps) to land at the same 4.24 as Vorbis, and still trailed AAC. The Broadcom result used a pre-final Opus bitstream, which is why it disagrees with the later tests.
5. Transparency Thresholds: Where Extra Bits Stop Mattering
“Transparent” has a precise meaning in audio coding: trained listeners in a controlled blind test cannot reliably tell the coded version from the original. Deezer Research found that Opus at 160 kbps met that bar under stress conditions (difference p = 0.747, strict equivalence p = 0.024), while 128 kbps did not once the hardest samples and sharpest ears were isolated (p = 0.030) (Deezer Research, Martin and Hennequin, arXiv 2026). In the first phase, with all 21 listeners, 128 kbps looked indistinguishable (p = 0.456), which shows how much the answer depends on who is listening.
The study used ITU-R BS.1116-3, a method in which listeners hear the original, then two hidden versions (one coded, one the original again) and grade the difference. A “difference grade” of 0 means no audible difference; the equivalence margin was plus or minus 0.25. Two caveats matter for streamers: the test covers Opus and music only, and it is a preprint under review. The broadcast surround figures in the table show the same logic at larger scale, with a safe bitrate well above the average-case one.
| Metric | Value | Source |
|---|---|---|
| Phase 1 mean difference grade vs uncompressed audio (0 = no audible difference): Opus at 32 vs 64 vs 128 kbps | -1.661 vs -0.438 vs -0.082 | Deezer Research, Martin and Hennequin, arXiv 2610.20246 (October 2026) |
| Phase 1, Opus at 128 kbps: difference test vs strict equivalence test | p = 0.456 (no detectable difference) vs p = 0.064 (equivalence not shown) | Deezer Research, arXiv 2610.20246 (2026) |
| Phase 2 stress test (10 most critical listeners, 8 hardest samples): difference tests at 96 vs 128 vs 160 kbps | p = 0.045 vs p = 0.030 vs p = 0.747 | Deezer Research, arXiv 2610.20246 (2026) |
| Phase 2, Opus at 160 kbps: strict equivalence test (margin plus or minus 0.25 grade) and mean difference grade | p = 0.024; -0.035 | Deezer Research, arXiv 2610.20246 (2026) |
| Listeners recruited vs qualified after screening; music excerpts | 26 vs 21; 23 excerpts | Deezer Research, arXiv 2610.20246 (2026) |
| 5.1 surround bitrate the EBU calls a safe choice for Excellent quality regardless of content | 448 kbit/s | EBU, Tech 3324 multichannel codec evaluations (September 2007) |
| HE-AAC 5.1 surround at 160 kbit/s or more | Excellent on average, applause only Fair | EBU, Tech 3324 (September 2007) |
| MUSHRA test rule: exclude a listener who rates the hidden original below 90 on more than this share of items | 15% | ITU-R, Recommendation BS.1534-3 (October 2015) |
Context note: Most recent available data for surround: 2007. MUSHRA (BS.1534) is the method for intermediate quality, scored 0 to 100 with anchors low-pass filtered at 3.5 kHz and 7 kHz, while BS.1116 is reserved for impairments so small that only rigorous conditions reveal them. A codec judged “Excellent” in MUSHRA is not automatically transparent in BS.1116 terms.
6. Speech-First Streams and the Listener’s Reference Points
Most live streams are talk with music underneath, and speech needs far fewer bits than music. A 2011 Google listening test rated fullband Opus at 32 kbps at 98.13 out of 100 on Mandarin speech, above G.719 at the same rate (93.43) (IETF, draft-ietf-codec-results-03 2013). Most recent available data: 2013.
The consumer services set the ear’s reference point. Spotify streams podcasts at about 96 kbit/s in its apps, and podcast feed guidance in the table tops out at 128 kbps for mono. A talk stream sent at 128 to 160 kbps is already above what listeners hear from the big podcast apps; a music stream at 128 kbps sits below Spotify’s High tier and far below its Very high tier.
| Metric | Value | Source |
|---|---|---|
| Google Mandarin speech test (21 listeners after screening, 0 to 100 scale): Opus fullband at 32 kbps vs G.719 at 32 kbps vs Opus wideband at 20 kbps | 98.13 vs 93.43 vs 81.59 | IETF, draft-ietf-codec-results-03 (May 2013) |
| Spotify podcast streaming bitrate: mobile and desktop apps vs web player | About 96 kbit/s vs 128 kbit/s | Spotify Support, Audio quality (2026) |
| Spotify music tiers: Low / Normal / High / Very high | About 24 / 96 / 160 / 320 kbit/s | Spotify Support, Audio quality (2026) |
| Spotify web player music: Free vs Premium | AAC 128 kbit/s vs AAC 256 kbit/s | Spotify Support, Audio quality (2026) |
| Spotify Lossless tier (Premium): format and maximum resolution | FLAC, up to 24-bit/44.1 kHz | Spotify Support, Audio quality (2026) |
| Apple Podcasts RSS feeds, AAC at 44.1 or 48 kHz: recommended mono vs stereo bitrate | 64 to 128 kbps vs 128 to 256 kbps | Apple, Apple Podcasts for Creators: Audio requirements (2026) |
| Apple Podcasts MP3 minimum: mono vs stereo | 32 kbps vs 64 kbps at 44.1 kHz | Apple, Apple Podcasts for Creators: Audio requirements (2026) |
Context note: on phones and tablets, choosing Low quality in Spotify also drops podcasts to about 24 kbit/s, a quarter of the default 96 kbit/s, which is the floor many listeners already accept for talk.
7. Headroom and Loudness Before the Encoder
Lossy encoders change waveforms, and peaks can come out higher than they went in. EBU guidance therefore places the final limiter before a lossy encoder at -2 dBTP, a full decibel below the -1 dBTP production ceiling (EBU, Tech 3344 v2.1 2016). “dBTP” means decibels true peak, a peak reading that includes the overshoots between digital samples. The same document warns that the limiter may need to go lower when bitrates are low, without giving a number.
For a live mix this is the most actionable figure in the article: a limiter set to 0 dBFS on the stream bus leaves the AAC or Opus encoder no room for its own overshoot. Loudness targets are a separate question, covered in depth in our streaming loudness normalization statistics; the EBU’s streaming supplement R 128 s2 (version 3.0, November 2023) recommends applying true-peak limiting in the playback device when a service uses a louder target.
| Metric | Value | Source |
|---|---|---|
| Maximum true-peak level for production material | -1 dBTP | EBU, Tech 3344 v2.1 (July 2016) |
| End-stage peak limiter setting in front of the lossy encoder | -2 dBTP (may need to be lower at low bitrates) | EBU, Tech 3344 v2.1 (July 2016) |
| Recommended maximum decoder output for AC-3 or E-AC-3 in RF mode | -4 dBTP | EBU, Tech 3344 v2.1 (July 2016) |
| Overshoot between the true-peak reading and the analogue output level | Less than 1 dB (typically a few tenths of a dB) | EBU, Tech 3344 v2.1 (July 2016) |
| Apple Podcasts true-peak ceiling | No higher than -1 dBFS | Apple, Apple Podcasts for Creators: Audio requirements (2026) |
| Apple HLS: loudness or dynamic range control metadata | Should be present unless all content, ads included, is encoded at the same level (requirement 2.18) | Apple, HLS Authoring Specification (2026) |
Context note: Most recent available data for codec headroom: 2016. None of the live ingest pages read for section 1 (YouTube Live, Facebook Live, Kick) publishes a loudness target, so the broadcast and podcast figures above are the closest documented reference.
Summary: Livestream Audio Bitrate and Encoding Settings by the Numbers
| Metric | Value | Source |
|---|---|---|
| Lowest Opus bitrate to pass strict transparency for music (BS.1116-3, equivalence p) | 160 kbps (p = 0.024) | Deezer Research, arXiv 2026 |
| Opus at 128 kbps vs uncompressed, 10 most critical listeners | Detected as degraded (p = 0.030) | Deezer Research, arXiv 2026 |
| YouTube Live stereo audio | 128 kbps at 44.1 kHz | Google, YouTube Help 2026 |
| Facebook Live audio bitrate: preferred vs maximum | 128 vs 256 kbps | Meta for Developers 2026 |
| 128 kbps audio as a share of Facebook Live’s 1080p60 video range | About 1.4% to 2.8% | Meta for Developers 2026 (our math) |
| Kick audio limit | Stereo, maximum 48 kHz | Kick Help Center 2026 |
| Spotify music tiers: High vs Very high | About 160 vs 320 kbit/s | Spotify Support 2026 |
| OBS Studio default audio bitrate and sample rate | 160 kbps, 48,000 Hz | OBS Project, source code 2026 |
| Apple HLS stereo AAC range | 32 to 160 kbit/s | Apple, HLS Authoring Specification 2026 |
| Opus recommended: fullband speech vs stereo music | 28 to 40 vs 64 to 128 kbit/s | IETF, RFC 6716 2012 |
| Codecs mandatory in every WebRTC endpoint | Opus and G.711 | IETF, RFC 7874 2016 |
| Score at a 96 kbps target, Opus vs Apple AAC (1 to 5) | 4.65 vs 4.40 | Hydrogenaudio test 2014 |
| Actual bitrate at that target, MP3 vs Opus | 135.5 vs 106.7 kbps | Hydrogenaudio test 2014 |
| 128 kbps audio as a share of YouTube Live’s 360p minimum video bitrate | 32% (128 / 400) | Google, YouTube Help 2026 (our math) |
| Spotify podcast bitrate in its apps | About 96 kbit/s | Spotify Support 2026 |
| Apple Podcasts recommended AAC stereo bitrate | 128 to 256 kbps | Apple Podcasts for Creators 2026 |
| Final limiter before a lossy encoder | -2 dBTP | EBU Tech 3344, 2016 |
Methodology and Sources
We read every figure in this roundup in the primary document during November 2026. Percentages in section 2 are our own arithmetic from published bitrates, with the formula shown next to each value.
- Deezer Research, Benjamin Martin and Romain Hennequin, Evaluating perceptual transparency of lossy music audio compression with Opus codec, arXiv preprint, October 2026.
- Google, YouTube Help: Choose live encoder settings, bitrates, and resolutions, 2026.
- Google, YouTube Help: Recommended upload encoding settings, 2026.
- Meta for Developers, Live Video API reference, 2026.
- Kick Help Center, How to stream on Kick, 2026.
- OBS Project, rtmp-services services.json, OBS Studio source code (default settings), and OBS Studio 30.0 release notes.
- IETF, RFC 6716 Opus (2012), RFC 7874 WebRTC audio codec requirements (2016), RFC 9725 WHIP (2025), and draft-ietf-codec-results-03 (2013).
- Apple, HLS Authoring Specification for Apple Devices and Apple Podcasts for Creators: Audio requirements, 2026.
- Hydrogenaudio community, Public multiformat listening test results, July 2014.
- EBU, Tech 3324: EBU evaluations of multichannel audio codecs (2007), Tech 3344 v2.1 (2016), and R 128 s2 version 3.0 (2023).
- ITU-R, Recommendation BS.1534-3 (2015) and Recommendation BS.1116-3 (2015).
- Spotify Support, Audio quality, 2026.
- Data watch: Twitch’s Broadcasting Guidelines page could not be rendered for reading, so the only Twitch number used is OBS’s preset ceiling (320 kbps). At least one third-party encoder guide (Switchboard) lists 160 kbps as the AAC maximum for Twitch, half the OBS preset; we could not confirm either figure against Twitch itself, so treat Twitch audio limits as unverified until checked on the live page. Kick publishes no audio bitrate. The Deezer study is a preprint under review for ICASSP 2027, its listeners were Deezer employees, and it tested Opus with music only, so its 160 kbps threshold should not be transferred to AAC or to speech. The Hydrogenaudio results page describes MP3’s extra bitrate both as “30kbps more” and as “29% more”; we report only the measured averages. EBU Tech 3324 cites over 14,000 scores from about 150 subjects in its conclusions, which does not match its per-lab tables (87 and 71 accepted listeners), so we do not use a listener total from it. Listening-test data in sections 4 and 5 is mostly older than three years (2007 to 2014), with the Deezer study the only recent exception. YouTube Live recommends 44.1 kHz for stereo while its upload guidance, Kick and OBS use 48 kHz.
Last updated: November 6, 2026. We update this roundup quarterly, and the next refresh is expected when the ICASSP 2027 proceedings publish the peer-reviewed version of the Deezer Research transparency study.