HAF - Home Audio Fidelity (Room Correction / convolution filter creation)

It sounds fantastic, no doubt about that… I was just curious

1 Like

Yeah whatever magic Thierry is doing is fantastic. I need to post more on my results.

All looks good, but here you only visualise 2 filters out of the 4.
The red curve is room/speakers correction of left channel
The green curve is reverb correction : using data of right channel, “fuzzed” with some random phase, amplitude correction to match “natural” reverberation, added to left channel
If you want to see the room/speaker correction of the right channel, upload the second channel of the HAF_44_HR.wav file.
It is easier to see the correction in logarithmic scale (here it’s linear and hard to read), start at 20Hz, vertical axis 100-120dB.

1 Like

I found this article interesting http://www.acousticfrontiers.com/2013127speaker-off-axis-response-psychoacoustic-and-subjective-importance/

Both on and off axis response must be similar in shape. If this is true then this means the early reflected and late reflected sound as heard in room will be similar in spectral balance to the direct sound.
Smoothness of the on and off axis curves – i…e absence of resonances – leads to higher subjective preference scores

This is exactly the point and the strength of the HAF filter : correct direct (on axis) and indirect (off axis, reverberated) responses + resonances of course.

Spoiler: improved “personalised” HAF X-talk cancellation filter is currently under beta testing on the French Hifi forum. I don’t know when it will be officially launched, I hope soon, because it’s really really good.

8 Likes

Finally got the chance to do my REW measurements and send them off to Thierry who has confirmed they are good to go. Have supplied a Glass Animals’ track as my test track.

Looking forward to hearing the results.

By the way using REW I equalised and saved the left/right channels from my position 1 (sweet spot) and loaded them into Roon. Definitely has helped remove some bass boom I was getting but not yet convinced the SQ has improved. I’ve lost some dimensionality to the music.

1 Like

You need several measurement spots to get good results, but expect to lose some “life” and “dynamic” when room correcting. Almost everyone feels that way, and the reason is that some sub-100 frequencies will be lowered, which gives the indication that you lost something. In fact, you will probably gain dynamic, but lose some volume, and it takes a little time before you realize how much of an improvement it is.

Btw, feel free to follow my REW guide here: A guide how to do room correction and use it in Roon and then you can compare that with what Thierry gives you. Thats what I did, and the HAF filters provided a SQ improvement at least for me.

If what you get from Thierry sounds worse, then give him a chance to fix it :wink:

4 Likes

When will this be available? Sign me up! :slight_smile:

I think its available now in beta, but you need to measure your head which feels a little medieval :slight_smile:

Ask Thierry about it, I will try it the next time I get new filters (need to buy some more acoustic treatment first)!

Thanks Magnus, I’ll send him an email. Measuring my head? huh?

Makes sense… my head is a massive echo chamber…

7 Likes

I’ve now received back from Thierry a couple of test tracks with and without crosswalk reduction. He has applied a neutral profile and he has warned me that they may sound light on bass. Will try them out this evening.

Amazingly quick response from Thierry!

1 Like

@Rik_Carter Remember you can always apply Roon DSP on top of HAF filter to adjust your profile. For example if you want more bass you can set a Parametric EQ / Low Shelf filter / Frequency 100Hz / Gain 3dB / Q=0.1

1 Like

The trick is to build a simple geometric “model” of your head based on 3 measurements, see example below.
The X-Talk filter is then built considering:

  • the room/speakers response measured with REW, correcting tone and phase for direct field and tone of the reverberated field (already taken in to consideration in current HAF version)
  • the distance of your head to the speakers and between speakers (already in current HAF version)
  • your head shape (“generic” head in current HAF version, “personalised head” in the new Beta version).
    The new Beta version is to my experience as good as the previous one at the sweet spot, slightly better on most “acoustic” recordings, and mind-blowing on some recordings with high binaural content. It’s getting into the direction of what can be achieved with BACCH filtering (except that the price is between 30 to 300 cheaper depending on the version of BACCH you choose, and it’s far easier to implement).

So you keep the mic next to you ears? Left ear and right ear are the sweetspot?

So a preface to my upcoming comment: I have played around with DRC a fair bit, having used Dirac, built my own convolution filters with REW - used in both Roon and HQP, and played with rePhase. As an engineer, I think I get most of the principles around the science. I have been excited to try the new HAF filters once my new setup is finalized - I don’t want to waste the $ before that.

So that out of the way … the idea that adjusting the filters for your head geometry is going to make an audible difference is going off the audiophile deep end imo. Unless you have your head in a vice grip that has been fixed to the exact same listening position as when you took the measurements, any movement of your head 1 or more inches in any direction is going to overwhelm anything related to your head geometry.

With all due respect, you are a case study in confirmation bias and/or have a close/working relationship with Thierry …

When you take REW measurements, there is only the mic at the sweet spot, stand away.
Don’t move the mic between left speaker and right speaker measurements.

I’ve now played through the original test track and the two modified tracks supplied by Thierry. The track is Glass Animals’ ‘Flip’ track which I’ve found a challenging track on every system I’ve ever tried, even my headphone setup. The most challenging is the final 1mins as there is a LOT of overlaid sounds on top of each other most of them in the same region as the vocals. This leads to a flattening wall of sound where everything sounds muddled together.

I noticed that the test tracks are quieter and Thierry confirmed that I should compensate by about +6dB on the test tracks to get them back to the same volume as the original track.

I’ve listened to all three tracks, three times (yes, I’m starting to get fed-up of listening to it). To my ears, the Crosstalk Reduction filter applied to the song is doing a marvellous job in that last 1 minute of the track. The other HAF is also tidying up that section of track but I just prefer the expansiveness of sound with the crosstalk reduction filter - I’ve no idea why that should be but there you go.

So, I’ve asked Thierry to go ahead and bill me for the crosstalk reduction filters. I’m very much looking forward to trying them out with the rest of my library :smiley:

Only question I’ve asked Thierry is whether he applies a bit of bass boost to his filters, or leave them neutral and I use the PEQ in Roon (as suggested by Eiffel), or should I just leave it for now and give myself some time to adjust to the filters with a neutral profile?

1 Like

We have similar background :slight_smile:

Head geometry and listening position/speakers position geometry are only of relevance and significance for crosstalk reduction. REW/Rephase/DIrac do not address this.

There’s nothing “idiophile” at all about it: cross-talk cancellation DSP have been around for years.
The simple idea is that your left ear should receive what comes from the left channel, and nothing from the right channel (and vice versa)
The trick is to compute a DSP that is preserving the signal from the left speaker and “cancelling” the right speaker at the left ear position, and vice versa. This is done by a matrix filter where data from the right channel is delayed, inverted, reshaped in frequency, and injected into the left channel in order to “cancel” what comes from the right channel to the left ear.
Why delayed: in a typical configuration your left ear is farther from the right speaker than the left. This delay depends on your sitting position relative to the speakers and distance between your ears, that’s why they need to be measured.
Inverted: so that the sound waves cancel and do not add (NB: if you edit your HAF filter you can see the impulse response of the x-talk is actually 180° out of phase with the room correction filter)
Reshaped in frequency: this is more tricky. Most likely your left ear is not directly pointing at the right speaker. It will hear whatever contribution the right speaker has on the reverberated field (this is the core of the HAF room correction) AND what will be diffracted by your head, which is frequency dependent (cf. https://search.yahoo.com/search?ei=utf-8&fr=aaplw&p=diffraction+sound+waves) . Typically, with my 23cm high head, frequencies below 1700Hz are diffracted (and heard directly by my ear), and the frequencies above are not diffracted and do not reach my hears directly. If you have a shorter head (let’s say 20cm), this transition frequency is 2000Hz. The X-talk cancellation filter has therefore to match the frequency “response” of your head as it is diffracting the sound waves coming from the right speaker. And this response is quite dependant on your head geometry, both vertically and horizontally. That’s why there is a need to measure it to “tune” the filter. With typical head sizes, diffraction occur in the 1.5k-2.5kHz area, this is incidentally where the ear is the most sensitive.
Regarding the size of the sweet spot where the X-talk works well: I’d say in my case it’s a 30cm radius “bubble” around the sweet spot. At greater distance, this is normal stereo listening with room correction (the one you have with the standard HAF filter).

I hope the technicalities above will convince you there is nothing idiophile about the approach of X_talk pushed by HAF. Regarding confirmation bias: if it would be only me to nod and clap hands, I would doubt of myself. Fortunately, there are lots of positive testimonies out there, including some from audiophiles that are far more experienced and critical than me.

Regarding my relationship with Thierry: I’ve said it on another post, but for the record I am not affiliated with him in any way. I paid for my filters. As a DSP-savvy person and audiophile I am enthusiastic about the technology and I am on this forum only to share my experience and explain DRC in general and HAF in particular. Until there is a new solution out there that is outperforming it !

6 Likes

Hey @alec_eiffel I can see where @nquery was coming from. Usually when someone is on a forum so enthusiastic about something it’s because they have a vested interest. Heck we live in the age of paid for reviews. From what you’re saying you’re just enthusiastic and keen to share your experiences.

We need more of that - “audiophiles” seem to be a solitary bunch. Listening to music is a personal experience. So we go into our bubble. Room calls this a community - not a forum. Forums are where you go to call other people stupid for having a different brand car / camera / whatever.

Thank you for sharing your experiences and this thread. It’s good community spirit and not selling. We need more sharing and community :slight_smile:

6 Likes

Do you have some references/white papers on how head height impacts diffraction? Obviously it does at some small empirical level, not arguing that. But to suggest that a 3 cm difference in head height can be consistently controlled for at that level seems dubious to me. Does the specific diffraction frequency change when I move around the room? What if I turn or angle my head 3 cm to the left - won’t that impact the results just as much? If your head remains in the same exact position, +/- a few cms, or better in a head vice grip, then I suppose it might work … but your 30 cm bubble seems generous. I will remain a skeptic for the time being …

edit: I did some searches and did some brief reading on head related frequency transfer and such … I am not skeptical of the science but still skeptical that it would have any audible impact in a non-controlled environment. But I am open to being proven wrong and may even try it once I get things sorted with my setup.