Old text to speech voices: Loquendo, Sam and today

If you grew up online in the 2000s, computer speech meant one of two engines. In English, it was Microsoft Sam, the default male voice in Windows 2000 and Windows XP, the one behind countless early YouTube skits. In Spanish, it was Loquendo, an Italian company whose voice Jorge narrated thousands of videos between 2006 and the early 2010s.

Both were clear, a little metallic and flat. That flaw became the signature: a deadpan robot reading something absurd was half the joke.

Loquendo was a speech technology company in Turin, Italy: text to speech, speech recognition and speaker verification. It began as a research group in the mid-1970s inside CSELT, the Italian phone company's lab, and became a standalone Telecom Italia company on February 1, 2001.

Its text to speech engine went into call centers, GPS units and phones, and individuals could buy its voices through reader programs like TextAloud. That consumer route is how Jorge, a Castilian Spanish male voice, ended up narrating GTA: San Andreas parodies, creepypastas and commented gameplays across Spain and Latin America.

Not legally. Neither voice was ever free software you could pass around. Loquendo voices were sold under license, one per voice and per program, and the brand stopped being sold separately after 2011. Microsoft Sam was part of Windows itself and was dropped as the default once Vista shipped. What circulates now as "Loquendo portable" or "Sam TTS download" is either an unlicensed copy or an unknown installer.

That's exactly the kind of download that brings infections. In a National University of Singapore study commissioned by Microsoft, 100% of the sites tested that offered pirated software exposed users to security risks, and trojans were the most common malware found in the downloaded copies (51% of infections).

• Installers that tell you to turn off your antivirus "so the crack works".
• Executables hidden inside a password-protected .zip or .rar.
• Programs from 2008 that won't install cleanly on current Windows and ask for "patches".

First figure out what you actually want: the exact robot sound of those old videos, or a narrator who reads your script clearly. Nothing does both.

What you can recreate from those videos is the tone: a serious, even narrator who reads anything with the same calm. This is how to do it on Voizum:

1. 1Open Voices and filter by English. For something close to Sam, try a deeper American male voice; for a newsreader feel, look at the narrator and news categories. Play a few samples before choosing.
2. 2In the generator, move Expressiveness toward "Restrained". The voice lets go less and sounds more neutral, like an announcer reading without acting.
3. 3Set the Speed. It goes from 0.5× to 2×. Those old videos often felt a little fast, so try nudging it above 1×.
4. 4Add pauses where you want a dramatic beat: <#1.5#> between two sentences adds a second and a half, up to 10 seconds.
5. 5Amounts and dates read fine on their own; acronyms are worth spelling the way you want them said. If "GTA" might come out wrong, write "G T A".
6. 6Generate a short test first. Once the tone works, generate the full script, download the MP3 and drop it into your editor.

Loquendo and similar engines of the 2000s worked like a collage. A voice actor recorded thousands of sentences over many hours; the software cut them into tiny pieces (syllables, transitions between sounds) and, to read your text, picked the pieces that fit and stitched them together. This is called unit selection synthesis.

That's why they sounded the way they did. The joins were audible, and the melody of a sentence came from fixed rules, not from what the sentence meant. A question and a statement came out almost the same.

A modern AI voice doesn't stitch anything. It has learned from a huge amount of recorded speech how a person sounds while reading, and it generates each sentence as a whole based on what it says: where the stress lands, whether the end drops for a statement or rises for a question. No seams, and the rhythm shifts from one sentence to the next.

For a 10-minute video, plan on roughly 10,000 characters of script, then fine-tune with the speed setting. Each voice has its own pace, and that moves the length more than the text does.

On Voizum, every 1,000 characters cost 60 credits, with no subscription, and purchased credits don't expire. Sign in with Google and you get 600 free credits each month, about 10 minutes of audio: enough to try several voices on your own script before paying anything.

Using an AI voice to narrate your videos, including commercial and monetized ones, is allowed. Using it to pass yourself off as someone else, or for illegal content, is not.

• Don't present your video as made with Loquendo or Microsoft Sam, or use their logos: those are other companies' trademarks.
• Don't try to reproduce the voice of a known YouTuber or announcer: imitating a real person without permission isn't allowed.
• Don't download "free Loquendo" or "Sam TTS" installers from random sites. Remember the Singapore study above.