How NPR webifies its programming — and you can, too

By Rob Holt and Chris Mandra

Nobody in public radio has encoded and streamed as much audio on the Internet — or had to automate the handling of such a large volume of material — as the staff at NPR Online. What advice do they have for stations that are new to streaming, or just thinking about starting? The writers are Rob Holt, webmaster of NPR Online, and Chris Mandra, production supervisor.

The statistics are clear: the time to webcast is now. There are more than 14,000 radio stations on the Web right now, building the interactive future of radio through the Internet medium. According to BRS Media (an e-commerce company), more than 3,500 of these stations webcast their on-air signal live. That’s up from 2261 stations last year, and only 56 in 1996, so the trend is clear. Of the 644-plus NPR member stations, more than 250 are on the Web, and dozens are currently streaming.

Webcasting extends a radio station’s reach from local to global, cultivating new audiences in a contemporary environment. For example, because WKSU in northeast Ohio established an early presence on the ’Net, they became a pre-set option for users of the first (and still dominant) streaming audio system, Real Audio. An international audience quickly developed, and WKSU received e-mail from listeners in South Africa, Korea, the Netherlands, and the Andes Mountains of Colombia. The station even got pledges from around the globe.

So, if your station isn’t on the Web, what’s stopping you? With four easy steps, you can be webcasting, too.

Step 1: Deciding what to webcast. You’ve decided the time is now to begin webcasting. The first strategic decision is whether to offer a live stream, archived original programming, or both. All three options have advantages and disadvantages. NPR Online offers a hybrid choice for our online listeners: we stream special events live (e.g., the National Press Club speeches and breaking news such as the impeachment), and we offer on-demand streaming of archived programs daily, such as Morning Edition and All Things Considered, starting daily after their last feed for the West Coast. Once the programs are available online, they remain available in an online archive that dates back to 1995.

The advantage of archived programming is that the listener becomes the programmer, choosing the songs or stories they want to hear when they want to hear them (on-demand streaming). However, on-demand programming requires a little more thought than streaming programming live. To offer archives, you must make a few a decisions:

  • Will you offer individual segments as well as the entire show to your online audience?
  • Do you want to get rid of time-sensitive material, e.g., newscasts, forward promotion and underwriting credits?
  • Do you want to offer listeners a choice of multiple bit rates, suitable to the speed of their Internet connections?

Each of these archiving options adds considerable time and effort to the online production process (outlined in Step 3); therefore, it is important to weigh their value as you plan your streaming.

Step 2: Starting with good audio. With the right information and “tools,” streaming can be a seamless and easy process with vast rewards. There are three main steps to streaming your station’s signal. First, the audio (usually a mono feed) enters the encoding computer through the line input of the computer’s sound card. The encoding software then translates the audio from the sound card into a streaming format and sends it over the network to the audio server. (For archived or on-demand streaming, the audio could be encoded “on the fly” and saved to a file for later streaming.) Lastly, whether the stream is live or archived, the server sends the audio data over the Internet to the player software (the “client” software) on the listener’s computer. The most common streaming systems include RealAudio/ RealMedia from RealNetworks, Windows Media from Microsoft, QuickTime from Apple, and Liquid Audio. RealNetworks’ players are still used by more online listeners around the country than the others, but Microsoft’s software may catch up with it.

A tip: Beware of claims that any particular player software is compatible with all streaming audio file types. For example, while Windows Media Player can stream RealAudio, it ignores some of the commands that work with RealPlayer, resulting in frustrating experiences for some users.

Potentially, the most important factor in creating a high-quality stream is the audio you start with. You work hard to make your over-the-air audio sound good — don’t stop at the threshhold of the Internet. The better it sounds going in, the better it will sound coming out. At NPR Online, we experimented with several audio options, looking to find the best quality audio for online listeners. We determined that the best approach is to obtain a direct feed from the studios, which has the top-quality sound listeners expect from public radio, and send it right into the computer. Keep the amplitude as high as possible without clipping.

Another tip: Though it’s important to start with clear audio that will stand up to the punishment of encoding, there is no need to handle excess high-frequency information that is lost in the encoding process. At NPR Online, we record our web audio at 22.05 khz, which omits some high frequencies but requires only half of the disk space used by audio of compact disc quality. We also record only a monaural stream. This applies all of the data compression and encoding trickery to make one channel sound good, instead of struggling with two channels. This advantage becomes negligible for transmissions at higher bit rates (56 kbps and up), so we provide stereo for those users.

With a sampling rate of 22.05 khz, the only equalization we’ve found necessary is a simple low-pass filter that takes out frequencies above 10 khz.

Step 3: Encoding your streams. The streaming process may sound complicated, but it will work as long as you have set it up properly. Encoding has gotten easier—you can now do it on an ordinary desktop computer. In the past, it was necessary to encode separate files for each transmission speed you wanted to offer on the web—14.4 kbps, 28.8 Kbps, 56 kbps, or higher. Today, with RealNetworks’ G2 encoder, you can create a single encoded file that supports multiple bit rates.

If you handle a high volume of web audio, you’ll want to automate recording and encoding as much as possible. To help us handle the detail work of offering many segments a day, we developed software called Apointer. Stations can do much the same thing by using the batch file capabilities of Sonic Foundry’s Sound Forge software. Sound Forge has an excellent batch-encode/processing feature. On Unix machines, the cron facility can be used to automate recording. On Macs and PCs there are other facilities that can be used to accomplish the same results, but we’ve found Unix machines to be incredibly reliable when it comes to automation and timing.

For most shows we put online, we break the show into two hour-long WAV files and remove the newscasts, promotion and underwriting credits using Sound Forge. We then mark “regions” (the start and end points for segments). Our Apointer software takes our two large edited files and splits them into individual WAV files, based on the regions we’ve marked, encodes each file separately (in the various bit rates and streaming formats we’ve chosen), glues the encoded individual files back together, and creates metafiles that point to the individual segments within the larger sound file. Apointer then transfers all of these resulting files to their appropriate locations on our servers so that our online listeners can hear programs streamed to their desktops.

The value of automation is best illustrated by our experience during President Clinton’s impeachment hearings and trial. NPR Online’s live coverage sometimes ran up to 13 hours a day. Because we had figured out how to have technology assist us in so much of our production process, we were able to stream audio live and archive it within a short time of its airing. The task was further complicated by our decision to give listeners direct access to the remarks of any member of Congress who spoke. We had to mark the start and end points of every speech, no matter how brief, and we had make sure each file started where the previous file left off. If we had done this without technological assistance, we would probably still be working on it. Because we were able to do all of this work quickly, a listener could come home at the end of the day and hear the words of Rep. Maxine Waters (D-Calif.) or Rep. Lindsey Graham (R- S.C.). Or, they could listen again to the whole day, or any part thereof. This just scratches the surface of what stations can offer web visitors.

Another option: Because we don’t stream programs live, until they finish airing on the West Coast, our “live” NPR Online stream should be called “pseudo-live.” We use a proprietary program called Audiolocker to automate this pseudo-live, 24-hour stream. The software enables us to play WAV files on a schedule and feed the audio into a “live” encoding computer. We tell Audiolocker that we want to loop Morning Edition between noon and 6 p.m. Eastern time, Talk of the Nation between 6 p.m. and 9 p.m., and then All Things Considered between 9 p.m. and midnight. Member stations, which may likewise want to create a pseudo-live stream, will soon be able to get a copy of Audiolocker, for Windows computers, from NPR’s members-only web site.

The software lets us loop files of unknown length. This is important because we do not know exactly how long a program will be from one day to the next after we edit out newscasts, forward promotions and underwriting credits. Audiolocker allows us to have files playing 24 hours a day. We simply copy the un-encoded WAV files to the Audiolocker server as part of the production process.

Step 4: Arranging to serve your sound. You will need a fast Internet connection and a powerful server to be able to handle many simultaneous audio streamers. A typical station should start with 60 or 100 simultaneous connections and expect to gradually increase capacity. Most stations licensed to universities will be able to get these services on campus.

Many other stations are having their audio served to the Web by partnering with local Internet Service Providers (ISPs), giving underwriting credits in exchange for web services.

Though stations can buy web servers of their own and do the job themselves, many prefer to hire it out for two basic reasons: maintenance expertise and bandwidth.

The first reason is that maintaining servers can be labor-intensive and expensive. Servers go down and log files fill up. Keeping up a server can be a 24-hour-a-day job. (At NPR Online, we have someone on call 24 hours a day to take care of web and audio server problems.) Stations can benefit from an ISP’s expertise.

Another reason to consider partnering with a local ISP is to give your web site a higher-capacity server and Internet connection so that it can handle growing usage and higher peak demands. NPR Online houses its audio servers off-site at a co-location facility that provides us with much higher bandwidth than we would have in our building.

Assembling the infrastructure needed to serve a decent number of simultaneous listeners can be quite costly. The average T-1 line runs at 1.54 megabits per second, which can support essentially 100 simultaneous encodes at 15 kpbs/second. If you have a T-1 and you get that much audio traffic to your site, you’ll notice that you will not be able to do much else over that connection. That means that your co-workers will not be able to access the Internet (and possibly your LAN) well and, in cases where the LAN and the phone system share the same T-1, you may have trouble making or receiving phone calls. Also, many ISPs have streaming software licenses for more streams than most of our member stations/program producers do. Most local ISPs are eager for creating relationships in the community and would be willing to work out in-kind trades.

Wait wait, what was that goal again?

In the midst of all the talks of bitstreams and baud rates, you must never lose sight of your goal: the more you use computers to assist your online production, the faster you can put valuable information on the Internet, the greater the public service you provide, and the larger the audiences you can attract.

If you figure out a better way of doing this, even if it is radically different, don’t be afraid to implement it. Be ever vigilant in your search for new ways to improve the online production process.

We are starting to see the development of a critical mass of public radio listening on the Internet. At any given time, between 300 and 400 people simultaneously listen to about 80 different archived sound files on NPR Online. We get the same number of people listening to our 24-hour pseudo-live stream. While it may not be a lot of people all at once, 500,000 files are listened to every week at NPR. Combine that number with the traffic at all the stations and you have a substantial service that did not exist before 1995.

Not long from now we will be able to offer web visitors the option of directly searching our audio archives. If listeners want to search the impeachment files for the phrase “grand jury testimony” and find all references, they will be able to. In the future, we will use today’s nascent technology to offer more service, more excitement, and to more broadly fulfill public radio’s mission of providing top-quality news, information, and entertainment programming to listeners.


Despite broadcasters’ fears, web audio hasn’t immediately eaten radio, 1998.

NPR’s first web-only program: All Songs Considered.


Current‘s links to public broadcasting’s streaming audio sites.