本帖最后由 留得青山在 于 2023-3-17 09:25 编辑
机翻：
在发烧友圈子里流传着这样的消息：数字音乐保存的双音传输只是义务，但自由式在别的地方。例如，我们正在谈论阻挡不需要的镜像频率的滤波器。您可以在音频 PC 基础知识中阅读有关采样率和抗锯齿滤波器的信息。下面我从论坛中从Jussi Laako（HQPlayer开发人员）那里获得了更多见解。 衰减在我们的基础知识中，已经有很多关于音频 PC HQPlayer 过滤器属性的信息。过滤器描述取自 HQPlayer 手册，不幸的是非常技术性。例如，滤波器“poly-sinc-gauss-xl”说的是极高的衰减。我们想在下面看看这意味着什么。图片显示，可听频率响应必须受到限制。例如，对于 CD （44.1 kHz），可听范围最多只能达到 22.05 kHz。基础是奈奎斯特-香农采样定理，它应该只考虑采样率的一半（奈奎斯特频率）。 现在，这取决于滤波器质量，实际允许通过哪些频率（通带），从阻塞效应（截止）开始以及需要多长时间（过渡范围）直到低通滤波器发挥其全部效果（停止频率）。如果阻塞效果非常低，则可以说阻尼非常高。这是一件好事，因为镜子频率得到了有效的抑制。高衰减可减少噪声伪影，并提高重建精度。 同样重要的是截止时间何时开始。一些滤波器已经在可听频率范围内开始，这当然是不可取的。陡峭的滤波器也是有利的，它实现了非常窄的过渡带（过渡区域）。如果我们以44.1 kHz（CD）的输入速率为例，“poly-sinc-gauss-xl”滤波器的通带高达20 kHz，转换范围为250 Hz宽，在奈奎斯特频率fs/2处达到22.05 kHz的最大衰减。 这种高质量的滤波器被Jussi Laako指定为快速或尖锐的“滚降”或“截止”。 资料来源：https://community.roonlabs.com/t/which-hqp-filter-are-you-using/6061/3242 过滤器长度公众通过Chord HUGO M SCALER开始意识到过滤器长度，它使用1万次抽头的过滤器。顺便说一下，例如，HQPlayer 使用具有 16 万个抽头的过滤器“封闭式 16M”轻松超越了这个数字。由于Chord为长滤波器做了如此多的广告，许多人认为滤波器长度是一个质量标准。Jussi Laako指出，滤波器在时域和频域上始终是一种折衷。 如果滤波器在时域中变长（抽头越多），则频域中的影响会更陡峭，反之亦然。这是因为频率和时间有 1/x 关系。滤波器越长，时域中振铃或“拖尾”的风险就越大。例如，长滤波器不擅长瞬态再现。另一方面，如果滤波器变得太短，它可以将高频分量减少到“减慢”瞬变的程度。太短的滤波器也会产生间隙，从而损失重建精度并产生可听见的失真。 在数学上，同时创建一个在这两个领域都完美的过滤器是不可能的。只有高带宽，如DSD记录（至少DSD128，但最佳DSD256+），才能同时提供两者。但是，带宽受限的 PCM 存在此问题。 挑战在于产生一个在时域中尽可能短的滤波器，同时在频域中不留间隙。换句话说，尽可能接近数学上不可能的事情。 在某种程度上，Chord 和 MQA 都是对的，也是错的。Chord 代表超长滤波器，而 MQA 代表超短滤波器。 它们是完全相反的。 两者都只有平均阻尼。 HQPlayer 提供时域和频域的滤波器，但大多介于两者之间，具有不同的权重。以及正在进行的工作，以更接近不可能的事情。古典音乐没有频繁的高速瞬变，所以更长的滤波器是可以的。“poly-sinc-gauss-long”和“poly-sinc-gauss-xla”都特别适合古典音乐和声音。还有“poly-sinc-ext2”和“poly-sinc-ext3”。顺便说一下，就初始化时间而言，过滤器“poly-sinc-long”是最苛刻的过滤器。相反，较短的滤波器（如“多辛-高斯-短”）由于瞬态播放而更适合电子、爵士、布鲁斯、流行和摇滚。 最长的过滤器是“sinc-L”。过滤器“sinc-LI”和“sinc-Mx”的长度相同。当然，这并不意味着它们是最好的，而只是最长的。实际长度还取决于转换率。为了进行比较，这些被调整为“sinc-Ll”大致相当于Chord M缩放器，“sinc-Lm”大约相当于Chord Dave，“sinc-Ls”大约相当于Chord Hugo/Mojo。 资料来源：https://audiophilestyle.com/forums/topic/19715-hq-player/?do=findComment&comment=1232215,https://audiophilestyle.com/forums/topic/19715-hq-player/?do=findComment&comment=1232738
线性相位和最小相位滤波器一组poly-sync-lp（线性相位 – lp）滤波器改善了空间性。它们属于FIR滤波器，在时域中工作。推荐用于古典音乐和“在现实世界”录制的音乐（音乐厅）。其中一个亚种是AsymFIR滤波器，特别适用于爵士/蓝调。除非另有说明，否则大多数滤波器都是线性相位的。 一组多同步-mp（最小相位 – mp）滤波器改善了瞬态。它们属于MinPhaseFIR滤波器，特别适用于摇滚/流行/电子，以及在音乐录音室录制。隐式地，这也适用于 IIR 滤波器。 来源： https://audiophilestyle.com/forums/topic/19715-hq-player/?do=findComment&comment=1232581 校准过滤器通常，我们依靠录音室的技能。尽管模拟记录良好，但数字化效果不佳可能是由于ADC（模数转换）故障造成的。另请参阅ADC和DAC之间的音频PC。 - 每个ADC抗混叠/抽取滤波器都会在数据上留下其“指纹”。校准过滤器可以更改此指纹。
- 典型的ADC抗混叠/抽取滤波器非常不完善，会在数据中留下误差。这些错误可以使用切趾过滤器进行纠正。错误的数字化也可能来自母带制作软件工具。
- 由于奈奎斯特频率，校准滤波器在红皮书源速率（44.1 kHz）下最为重要。
卷积（用于房间和扬声器校正的卷积滤波器）HQPlayer 不仅提供了许多过滤器，还可以使用您自己的过滤器。例如，用于房间和扬声器校正。另请参阅： - 音频 PC 提供了哪些改善室内声学的可能性？
- 为什么音频 PC 可以优化扬声器？
虽然具有不同采样率的ZIP文件必须存储在Roon DSP中，但每个具有采样率的通道只有一个滤波器存储在HQPlayer中。使用“HF 扩展”选项，HQPlayer 自行计算所需的采样率。但是，最好将最高采样率作为卷积滤波器存储在 HQPlayer 中。例如，Acourate可用于创建高达384 kHz的高质量卷积滤波器。 总结高质量的滤波器，如HQPlayer中的滤波器，可以非常有效地抑制镜子频率。其他重要的滤波器标准包括哪些频率仍允许通过（通带），截止时间何时开始以及低通滤波器需要多长时间（过渡范围）才能发挥其全部效果（停止频率）。好的滤波器具有快速截止和极高的衰减。 长滤镜绝不是唯一的质量特征，但根据音乐流派和基本上在时间范围内，甚至会产生负面影响。长滤波器适用于古典和人声，短滤波器适用于基于瞬态的音乐，如摇滚和流行音乐。滤波器始终是时域和频域的折衷方案。理论就这么多。我真的很喜欢听“poly-sinc-gauss-xla”，带有具有极高衰减的切趾超长滤波器，尽管古典音乐几乎从不包括在内。但随着民谣和独立音乐的很多声音。 多同步-lp（线性相位 – lp）滤波器组改善了空间度，而多同步-mp（最小相位 – mp）滤波器组和隐式IIR滤波器组改善了瞬态。 校准滤波器可以校正不良的模数（ADC）转换。 如果使用卷积滤波器，请在 HQPlayer 中存储尽可能高的采样率。
机翻英文： Word has gotten around in audiophile circles that the bitidendic transmission of digital music preserves is only the obligation, but the freestyle lies somewhere else. For example, we are talking about filters that block out unwanted mirror frequencies. You can read about sample rates and anti-aliasing filters in the Audio PC Basics. Below I have taken some more insights from Jussi Laako (HQPlayer developer) from the forums. AttenuationIn our basics there is already a lot about the Audio PC HQPlayer filter properties. The filter descriptions were taken from the HQPlayer manual and are unfortunately very technical. For example, the filter "poly-sinc- gauss-xl" speaks of an extremely high attenuation. We want to take a look at what this means below. The picture shows that the audible frequency response must be limited. For example, with a CD (44.1 kHz), the audible range should only go up to a maximum of 22.05 kHz. The basis is the Nyquist-Shannon sampling theorem, which should only take into account half the sampling rate (Nyquist frequency). Now it depends on the filter quality, which frequencies are allowed through in practice (passband), from when the blocking effect (cutoff) begins and how long it takes (transition range) until the low-pass filter unfolds its full effect (stop frequency). If the blocking effect is very low, one speaks of an extremely high damping. This is a good thing because mirror frequencies are effectively suppressed. High attenuation results in less noise artifacts and also better reconstruction accuracy. Equally important is when the cutoff begins. Some filters already start in the audible frequency range, which of course is not desirable. A steep filter is also favorable, which achieves a very narrow transition band (transition area). If we take the input rate of 44.1 kHz (CD) as an example, the "poly-sinc- gauss-xl" filter has a pass band of up to 20 kHz, the transition range is 250 Hz wide and reaches the maximum attenuation at the Nyquist frequency fs/2 at 22.05 kHz. Filters of this high quality are designated by Jussi Laako as fast or sharp "rolloff" or "cutoff". Sources: https://community.roonlabs.com/t/which-hqp-filter-are-you-using/6061/3242 Filter lengthThe general public became aware of filter lengths with the Chord HUGO M SCALER, which uses filters with 1 million taps. By the way, a number that the HQPlayer, for example, easily surpasses with the filter "closed-form-16M" with 16 million taps. Because Chord has done so much advertising for long filters, many think that filter length is a quality criterion. Jussi Laako points out that filters are always a compromise in the time and frequency domain. If the filter gets longer in the time domain (more taps), the effect in the frequency domain is steeper and vice versa. That's because frequency and time have a 1/x relationship. The longer the filter, the greater the risk of ringing or "smearing" in the time domain. Long filters, for example, are not good at transient reproduction. On the other hand, if the filter becomes too short, it can reduce the high-frequency components to such an extent that it "slows down" the transients. A filter that is too short can also create gaps, so that the reconstruction accuracy is lost and audible distortions are created. Creating a filter that is perfect in both areas at the same time is mathematically impossible. Only high bandwidths, as with DSD recordings (at least DSD128, but optimal DSD256+), can offer both at the same time. However, bandwidth-constrained PCM has the problem. The challenge is to produce a filter that is as short as possible in the time domain and at the same time leaves no gaps in the frequency domain. In other words, to get as close as possible to the mathematically impossible. In a way, both Chord and MQA are right and wrong. Chord stands for super long filters, while MQA stands for super short filters. They are completely opposites. Both have only average damping. HQPlayer offers filters in both the time domain and the frequency domain, but mostly something in between, with different weights. And ongoing work to get even closer to the impossible. Classical music doesn't have frequent high-speed transients, so longer filters are fine. Both "poly-sinc-gauss-long" and "poly-sinc-gauss-xla" are particularly suitable for classical music and voices. Also "poly-sinc-ext2" and "poly-sinc-ext3". By the way, the filter "poly-sinc-long" is the most demanding filter in terms of initialization times. Conversely, shorter filters such as "poly-sinc- gauss-short" are better for electronic, jazz, blues, pop and rock due to transient playback. The longest filter is "sinc-L". The filters "sinc-LI" and "sinc-Mx" are the same length. Of course, this does not mean that they are the best, but only the longest. The actual length also depends on the conversion ratio. For comparison, these are tuned so that "sinc-Ll" is roughly equivalent to Chord M scaler, "sinc-Lm" is approximately equivalent to Chord Dave, and "sinc-Ls" is approximately equivalent to Chord Hugo/Mojo. Sources: https://audiophilestyle.com/forums/topic/19715-hq-player/?do=findComment&comment=1232215,https://audiophilestyle.com/forums/topic/19715-hq-player/?do=findComment&comment=1232738
Linear-phase and minimal-phase filtersThe group of poly-sync-lp (linear phase – lp) filters improve the spatiality. They belong to the FIR filters and work in the Time domain. Recommended for classical music and for music recorded "in the real world" (concert hall). A subspecies of this are the AsymFIR filters, which are particularly suitable for jazz/blues. Most filters are linear-phase unless otherwise noted. The group of poly-sync-mp (minimum phase – mp) filters improve the transients. They belong to the MinPhaseFIR filters and are particularly suitable for rock/pop/electronics, as well as recorded in the music studio. Implicitly, this also applies to IIR filters. Source: https://audiophilestyle.com/forums/topic/19715-hq-player/?do=findComment&comment=1232581 Calibration filterNormally, we rely on the skills of the recording studios. Despite good analog recordings, poor digitization can be due to faulty ADC (analog-to-digital conversion). See alsoAudio PC between ADC and DAC. - Each ADC anti-alias/decimation filter leaves its "fingerprint" on the data. Calibration filters can change this fingerprint.
- Typical ADC anti-aliases/decimation filters are quite imperfect and leave errors in the data. These errors can be corrected with an apodization filter. Faulty digitization can also come from the mastering software tools.
- Due to the Nyquist frequency, calibration filters are most important at Red Book source rates (44.1 kHz).
Convolution (convolution filter for room and speaker correction)The HQPlayer not only provides many filters, but can also use your own filters. For example, for room and speaker correction. See also: - What possibilities does an audio PC offer to improve room acoustics?
- Why can an audio PC optimize the speaker?
While a ZIP file with the different sampling rates must be stored in the Roon DSP, only one filter per channel with a sampling rate is stored in HQPlayer. With the "HF Expand" option, the HQPlayer calculates the required sampling rate itself. However, it is better to store the highest sampling rate as a convolution filter in HQPlayer. For example, Acourate can be used to create convolution filters in very high quality up to 384 kHz. SummaryHigh-quality filters such as in HQPlayer dampen the mirror frequencies very efficiently. Other important filter criteria are which frequencies are still allowed through (passband), when the cutoff kicks in and how long it takes (transition range) for the low-pass filter to unfold its full effect (stop frequency). Good filters have a fast cutoff and extremely high attenuation. Long filters are by no means the sole quality feature, but can even have a negative effect depending on the music genre and basically in the time range. Long filters are suitable for classical and voices, short filters for transient-based music such as rock and pop. Filters are always a compromise in the time and frequency domain. So much for the theory. I really like to hear "poly-sinc- gauss-xla" with the apodising extra long filter with extremely high attenuation, although classical music is almost never included. But with folk and indie a lot of voices. The group of poly-sync-lp (linear phase – lp) filters improve spatiality while the group of poly-sync-mp (minimum phase – mp) filters and implicit IIR filters improve transients. Calibration filters can correct poor analog-to-digital (ADC) conversions. If you use convolution filters, store the highest possible sampling rate in HQPlayer.
| 
楼主: 留得青山在
粤公网安备 44030602000598号 耳机大家坛、www.erji.net、网站LOGO图形均为注册商标 
IP卡
狗仔卡
楼主
显身卡
发表于 2023-3-17 07:56
