The Voice of Latino America: How a Bilingual AI Is Reshaping Spanish-Language Radio

The radio in the kitchen is doing what has always characterized the Latino neighborhoods of East Los Angeles on a weekend morning: switching. The DJ’s voice switches between Spanish and English, sometimes in the same sentence, just like a conversation between cousins during a family meal. The audience fully comprehends this, and the station, 96.3 La Mega, has been doing this since 2005. They have spent their entire lives switching codes.

They might not be aware that AI is increasingly present in the room with them, listening, analyzing, and occasionally responding.

Over sixty million Americans listen to Spanish-language radio, which has long been regarded as the fastest-growing format in American broadcasting. Additionally, it carries misinformation, a persistent issue that has been challenging to quantify and more difficult to resolve, in some markets and on specific frequencies. false information about health. false information about politics. During election cycles and public health emergencies, content created by foreign actors—including Russian outlets—finds its way into Spanish-language programming. For many years, journalists and watchdog groups have monitored this issue primarily by hand, listening to hours of radio, manually translating, and marking what they discovered. Every time, the volume exceeded the monitoring capacity.

One of the more direct attempts to alter that is a tool called VERDAD, which is the Spanish word for truth. The app, which was created by Wayne State University journalist Martina Guzman and released in 2025 with assistance from AI experts in California, employs AI to spot potentially deceptive content in Spanish-language radio broadcasts and translate it into English so that researchers and journalists can review it. Language, station, state, topic, and political inclination are among the filters available to users. It presents content for human review rather than rendering editorial decisions. Guzman has taken care to maintain this important distinction.

Fact-checking is only one aspect of the larger change taking place in Spanish-language radio. Sub-100 millisecond Spanish-English translation is now possible thanks to programs like Transync AI. This speed allows conversations between speakers of the two languages to flow naturally without the lag and pause that usually indicate someone is waiting for a machine to catch up. The question of where synthetic voice ends and real broadcasting begins is being raised by voice synthesis platforms such as ElevenLabs, which are producing AI voices in Spanish with increasingly complex regional accent variations. A multicultural radio broadcasting organization called COPEAM has been testing and documenting AI-assisted transcription, translation, and linguistic adaptation workflows, advancing AI integration from research to real-world production settings.

The conflict between efficiency and authenticity that permeates all of this is one that the radio industry itself has been negotiating in different forms for decades. Spanish-language radio has always carried a special weight, especially in places like Los Angeles, Miami, Houston, and Chicago. It is where advertisers reach a community with $3.4 trillion in collective spending power, where second-generation listeners find a cultural middle ground between English and Spanish, and where first-generation immigrants find news in their native tongue. No matter how technically sound an AI voice or an algorithm-generated segment sounds, the trust ingrained in that relationship does not automatically translate to it.

A more subdued question is whether the AI tools being created for this field are truly designed for it. With Spanish as a secondary focus and Latin American regional dialects frequently underrepresented, the majority of large language models and voice synthesis systems have been trained mainly on English-language data. When using Dominican or Chicano Spanish, a model trained on Castilian Spanish performs differently. This issue has been brought up explicitly by researchers and Latin American AI developers, who contend that the region requires AI models constructed from its own linguistic and cultural data rather than generalized solutions that have been retrofitted from elsewhere.

It’s difficult to ignore the obvious and largely unresolved problem of providing sixty-one million Spanish-speaking listeners with AI tools that truly comprehend their speech. The localized data pipelines that would enable the tools to function properly are not arriving as quickly as the tools. VERDAD is operating. Transync is operational. In the kitchen, the radio continues to play and switch between languages. It is still genuinely unclear who is listening and how well.