Skip to content

Releases

v0.3.8 — voice thinned at worst, never removed

Choose a tag to compare

@HandsomeWJ HandsomeWJ released this 06 Oct 12:27

Fork of alexeygrigorev/microboost by Alexey Grigorev, who wrote the original app.

0.3.8: your voice is thinned at worst, never removed

With a speech-heavy video, 0.3.7 could make you almost inaudible while it played: the suppressor's aggressive setting treats every band your voice does not clearly dominate as echo. 0.3.8 makes the talk detector more sensitive and, while it hears you, switches to a gentle prediction and caps every band at −12 dB, holding for half a second after you stop so word onsets survive.

Measured on the real room recording at Strength 50: voice gain 0.99 / 0.95 / 0.82 with your voice at +12 / 0 / −12 dB relative to the echo; speakers still 20 to 22 dB down while you are silent.

Download: microboost-windows-amd64.exe (Windows 10+, x64). Requires VB-CABLE, which the app offers to install on first run.

SHA-256: ef7b0161759003afa70e21bd59c6afddb3d824217929687e5dab480f6f5bd0ec

v0.3.7 — selectable reference output device

Choose a tag to compare

@HandsomeWJ HandsomeWJ released this 06 Oct 12:20

Fork of alexeygrigorev/microboost by Alexey Grigorev, who wrote the original app.

0.3.7: choose which output device is the reference

Realtek USB codecs expose separate "Speakers" and "Headphones" output endpoints. When Windows' default output is one and a video plays on the other, the app captured silence as its reference and suppressed nothing (status "Speakers: silent" while a video was audibly playing).

  • New Reference dropdown in the echo section: Windows default, or any output device. Pick the one the video actually plays on.
  • When following the default, the capture restarts automatically if Windows changes the default output.
  • If the reference stays silent for 3 s while the mic hears sound, the UI says so and names the device it is capturing.

Download: microboost-windows-amd64.exe (Windows 10+, x64). Requires VB-CABLE, which the app offers to install on first run.

SHA-256: 34eca5a2d5a70f1c99e0116b047788a2ddc29bc1cf120edd22f0589a962f2475

v0.3.6 — tuned on a real room recording

Choose a tag to compare

@HandsomeWJ HandsomeWJ released this 06 Oct 09:55

Fork of alexeygrigorev/microboost by Alexey Grigorev, who wrote the original app.

0.3.6: tuned on a real room

Analysis of a real echo_diag.wav (USB mic needing 50x, wired desktop speakers) showed the mic picking the speakers up mostly as 100–300 Hz bass that is only partly coherent with the signal (desk vibration or speaker distortion). A linear canceller cannot beat about 5 dB on that, so the per-band suppressor now carries the load:

  • The suppression factor is set by Strength (16× over-prediction at 40 dB, doubling every 10 dB) and no longer depends on the talk detector. Default 50 dB for new installs; existing users: move the slider to 50.
  • The talk detector looks at the speech band only (300 Hz up): 0% false alarms on the recording while silent.
  • Window relocation needs two agreeing lag estimates.

Measured on the recording at 50 dB: output 20 dB down with you silent (that is the mic's noise floor); talking 12 dB above the echo, speaker audio 8.5 dB down, voice gain 0.96, voice distortion −19 dB (slightly thinner bass while media plays).

Download: microboost-windows-amd64.exe (Windows 10+, x64). Requires VB-CABLE, which the app offers to install on first run.

SHA-256: 8464391dd65f6e36e362997493c12256e0414459de2a9e200f11ea65d28c4227

v0.3.5 — echo diagnostics

Choose a tag to compare

@HandsomeWJ HandsomeWJ released this 06 Oct 09:39

Fork of alexeygrigorev/microboost by Alexey Grigorev, who wrote the original app.

0.3.5: diagnostics

Real hardware still shows about 1 dB cancelled with echo detected after the timestamp alignment of 0.3.4. This build adds the instrumentation needed to tell why, instead of another guess:

  • A diagnostics line under the echo status: loopback device and format, envelope correlation, best lag (negative means the mic leads the reference), the filter window, max and typical cancellation, timing offset, and the share of reference samples that were unavailable.
  • A Record 10 s echo diagnostic button that saves the raw mic (left) and the speaker reference (right) to %APPDATA%\Microboost\echo_diag.wav for offline analysis.

Download: microboost-windows-amd64.exe (Windows 10+, x64). Requires VB-CABLE, which the app offers to install on first run.

SHA-256: d64e8646a39a27db4e8cca4e2569969656c3f7ce340f87ec760d821f544a2765

v0.3.4 — reference aligned by timestamps

Choose a tag to compare

@HandsomeWJ HandsomeWJ released this 06 Oct 09:23

Fork of alexeygrigorev/microboost by Alexey Grigorev, who wrote the original app.

0.3.4: fixes "cancelling 0 dB"

The status line on real hardware showed echo detected but 0 dB cancelled. Cause: the copy of the speaker audio was read from its buffer by position, and depending on buffer sizes and which stream started first it could sit behind the echo in the mic. A causal filter can only use reference samples that arrive before the echo, so it learned nothing.

The reference is now aligned to the mic by capture timestamps (both WASAPI streams stamp their buffers from the same performance counter), read 15 ms earlier than the mic so the echo always lags it, with clock drift between mic and sound card nudged out continuously. The status line now also shows the measured offset ("ref +N ms").

Download: microboost-windows-amd64.exe (Windows 10+, x64). Requires VB-CABLE, which the app offers to install on first run.

SHA-256: 0257a15df00fbe52081cd30bec0f73b63e3991e0995c37d9b693564bd9e936c6

v0.3.3 — your voice can no longer be muted

Choose a tag to compare

@HandsomeWJ HandsomeWJ released this 06 Oct 05:08

Fork of alexeygrigorev/microboost by Alexey Grigorev, who wrote the original app.

0.3.3: fixes "completely inaudible while a video plays"

0.3.2 had a deadlock: the detector that decides whether you are talking needed the canceller's measured performance to exceed 5 dB, but that measurement could only grow while the detector said you were silent. If you talked from the start it stayed near zero, the code fell back to muting the whole mic whenever speakers played, and you vanished.

0.3.3 removes that dependency and the broadband mute altogether. The leftover-echo estimate is now learned with minimum statistics, which talking cannot poison, and it keeps working even when the linear canceller cancels little. The only way to be attenuated is to be quieter than the leftover echo itself.

Test results: linear path, talk-over speaker leakage −37 dB, voice gain 1.00, voice distortion −58 dB, speakers-only −40 dB. Saturating speaker with your voice 6 dB below the echo: voice gain 1.00, speaker audio −11 dB.

Download: microboost-windows-amd64.exe (Windows 10+, x64). Requires VB-CABLE, which the app offers to install on first run.

SHA-256: b4006d8f48cd846f33b47579019b8d3f576be39a460a519d309f0ff17052f313

v0.3.2 — reduce speakers while you talk

Choose a tag to compare

@HandsomeWJ HandsomeWJ released this 06 Oct 04:55

Fork of alexeygrigorev/microboost by Alexey Grigorev, who wrote the original app.

0.3.2: speakers are reduced even while you talk

0.3.1 only cancelled the speaker audio while you talked and switched the extra suppression off, because one gain for the whole signal cannot cut one voice without cutting the other. 0.3.2 suppresses per frequency band instead: in each 5 ms frame, bands where leftover speaker audio is louder than your voice are attenuated, bands where your voice dominates are left alone. Your voice level is not changed.

Synthetic test results (speech fixture, mic noise): while talking over the speakers the speaker audio sits 39 dB below its raw level in the mic, your voice passes at gain 1.00 with −29 dB distortion (mild coloration); with you silent the output is 44 dB down. Latency is now ~11 ms.

Status line: "cancelling N dB · suppressing leftover up to M dB · you are talking / only echo".

Download: microboost-windows-amd64.exe (Windows 10+, x64). Requires VB-CABLE, which the app offers to install on first run.

SHA-256: c04605a58339cfdf4bd24e33d142aaefde5307a0c4510ca0a61be877b7b7b212

v0.3.1 — keep your own voice while suppressing speakers

Choose a tag to compare

@HandsomeWJ HandsomeWJ released this 06 Oct 04:40

Fork of alexeygrigorev/microboost by Alexey Grigorev, who wrote the original app.

0.3.1: your own voice is no longer ducked

0.3.0 cancelled the speaker audio correctly but then attenuated the whole mic by strength − cancelled dB whenever the speakers played, which cut your voice during talk-over. 0.3.1 ducks only while the mic holds nothing but leftover echo: when your voice is louder than the predicted residual and the mic's noise floor, ducking releases within 10 ms and stays off for 300 ms after you stop. The status line now says "you are talking" or "only echo".

Measured in the synthetic tests: with speakers playing and you silent the mic is ducked ~30 dB; while you talk over the speakers your voice passes at full level.

Download: microboost-windows-amd64.exe (Windows 10+, x64). Requires VB-CABLE, which the app offers to install on first run.

SHA-256: 5f9909e8698bd95e6817a74d8d20bdb402207b48c740f97ca8b577aee16495c7

v0.3.0 — speaker echo suppression

Choose a tag to compare

@HandsomeWJ HandsomeWJ released this 06 Oct 04:24

Fork of alexeygrigorev/microboost by Alexey Grigorev, who wrote the original app.

New in 0.3.0: speaker echo suppression

At 50x boost the mic also picks up whatever the speakers play (podcasts, videos, call audio). The app now captures the speaker mix via WASAPI loopback and removes it from the mic:

  • Adaptive cancellation (default): a partitioned frequency-domain adaptive filter learns the speaker→mic path (delay up to 1 s) and subtracts the echo, then ducks only the leftover. You can talk while something plays. Ducking engages only when echo is actually detected in the mic, so headphone users are left alone.
  • Adaptive off: the mic is attenuated by the chosen strength whenever the speakers play.
  • Live status line: speakers playing/silent, echo detected, dB cancelled, dB ducked.
  • Adds ~5 ms latency. Uses about 1% of one CPU core.
  • Cannot remove a person in the room, only sound the PC itself is playing.

Unit-tested with synthetic echo paths (41 dB cancellation, 150 ms delay relocation, double-talk, no-echo). Not yet exercised on real Windows audio hardware; please report what the status line shows if it misbehaves.

Download: microboost-windows-amd64.exe (Windows 10+, x64). Requires VB-CABLE, which the app offers to install on first run.

SHA-256: d67b162d6864897944372e73e2325b85036b2e63e7a9fb3254fbb5174fc89ca4

v0.2.0 — 50x boost, tray icon, auto-start

Choose a tag to compare

@HandsomeWJ HandsomeWJ released this 06 Sep 16:20

Fork of alexeygrigorev/microboost by Alexey Grigorev, who wrote the original app.

Changes in this fork

  • Boost cap raised from 5x to 50x (logarithmic slider; auto-calibrate can recommend up to 50x)
  • System tray: minimise/close hides to the tray and keeps boosting. Left-click reopens, right-click for Show/Quit
  • Settings section: auto-start boost on launch, start hidden in tray, launch with Windows (per-user Run key)
  • Cross-compile instructions for macOS via cargo-xwin

Download: microboost-windows-amd64.exe (Windows 10+, x64). Requires VB-CABLE, which the app offers to install on first run.

SHA-256: a192c0025335d95b198c97abb012bccb4498b2a9ec722868187882d73cbba785