diff --git a/raptor-indexer-csv/src/main.rs b/raptor-indexer-csv/src/main.rs
index b26dfcd9f..763a4c962 100644
--- a/raptor-indexer-csv/src/main.rs
+++ b/raptor-indexer-csv/src/main.rs
@@ -7,11 +7,10 @@ use std::path::{Path, PathBuf};
 use std::collections::{HashSet, BTreeMap};
 use std::io::{self, BufReader, BufRead};
 use std::fs::File;
-use std::iter;
 
 use csv::ReaderBuilder;
 use structopt::StructOpt;
-use raptor::{MetadataBuilder, DocIndex};
+use raptor::{MetadataBuilder, DocIndex, Tokenizer};
 use rocksdb::{SstFileWriter, EnvOptions, ColumnFamilyOptions};
 use unidecode::unidecode;
 
@@ -55,6 +54,30 @@ where P: AsRef<Path>,
     Ok(set)
 }
 
+fn insert_document_words<'a, I, A, B>(builder: &mut MetadataBuilder<A, B>, doc_index: u64, attr: u8, words: I)
+where A: io::Write,
+      B: io::Write,
+      I: IntoIterator<Item=(usize, &'a str)>,
+{
+    for (index, word) in words {
+        let doc_index = DocIndex {
+            document: doc_index,
+            attribute: attr,
+            attribute_index: index as u32,
+        };
+        // insert the exact representation
+        let word_lower = word.to_lowercase();
+
+        // and the unidecoded lowercased version
+        let word_unidecoded = unidecode(word).to_lowercase();
+        if word_lower != word_unidecoded {
+            builder.insert(word_unidecoded, doc_index);
+        }
+
+        builder.insert(word_lower, doc_index);
+    }
+}
+
 fn main() {
     let opt = Opt::from_args();
 
@@ -85,29 +108,13 @@ fn main() {
             Err(e) => { eprintln!("{:?}", e); errors += 1; continue },
         };
 
-        {
-            let title = iter::repeat(0).zip(product.title.split_whitespace()).filter(|&(_, w)| !common_words.contains(w)).enumerate();
-            let description = iter::repeat(1).zip(product.description.split_whitespace()).filter(|&(_, w)| !common_words.contains(w)).enumerate();
+        let title = Tokenizer::new(&product.title);
+        let title = title.iter().filter(|&(_, w)| !common_words.contains(w));
+        insert_document_words(&mut builder, product.id, 0, title);
 
-            let words = title.chain(description);
-            for (i, (attr, word)) in words {
-                let doc_index = DocIndex {
-                    document: product.id,
-                    attribute: attr,
-                    attribute_index: i as u32,
-                };
-                // insert the exact representation
-                let word_lower = word.to_lowercase();
-
-                // and the unidecoded lowercased version
-                let word_unidecoded = unidecode(word).to_lowercase();
-                if word_lower != word_unidecoded {
-                    builder.insert(word_unidecoded, doc_index);
-                }
-
-                builder.insert(word_lower, doc_index);
-            }
-        }
+        let description = Tokenizer::new(&product.description);
+        let description = description.iter().filter(|&(_, w)| !common_words.contains(w));
+        insert_document_words(&mut builder, product.id, 1, description);
 
         // TODO simplify this by using functions and
         //      use the MetadataBuilder internal BTreeMap ?
diff --git a/raptor/src/lib.rs b/raptor/src/lib.rs
index 999218169..122c1f69e 100644
--- a/raptor/src/lib.rs
+++ b/raptor/src/lib.rs
@@ -8,6 +8,7 @@ pub mod tokenizer;
 
 pub use self::metadata::{Metadata, MetadataBuilder};
 pub use self::rank::RankedStream;
+pub use self::tokenizer::Tokenizer;
 
 pub type DocumentId = u64;